Before this change, HlsMediaSource timelines had a period starting at the epoch.
For VOD streams the window position in the period was the program date time.

This change makes period and initial window positions match. For live streams
the window position advances as segments are removed, so its position in the
period is the difference between the initial program date time and the program
date time of the latest playlist.

This also makes it possible to insert ads in VOD HLS content with program date
time, as the period and window are now aligned.

Issue: #3865

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=187590948

AdsMediaSource must be top-level. Currently the (deprecated) ImaAdsMediaSource
can't be used because it prepares its contained AdsMediaSource as a child
source.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=186990882

r2.7.0

Allow the position to jump on receiving the first presentable input buffer,
if and only if the buffer timestamp differs significantly from what was
expected. This prevents a stuck buffering case for streams that are thought
to start at t=0, but actually start at t=large_value.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=186454715

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=186305613

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=186291489

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=186200840

Occasionally the player could transition from playing content to playing an ad
after IMA called playAd. The discontinuity triggered faking the content
position, and the fake position was passed to IMA when content resumed causing
the wrong ad group to be loaded. Fix this by only faking the position if the
player transitions before playAd.

Also fix the calculation of the expected ad group index for a postroll ad, and
wait for the player to transition back from ad to content before passing a
content progress update.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185994229

If an exception is thrown in an IMA callback it crashes the process with lots of
logging from WebView (including several stack traces, etc.). This change wraps
ImaAdsLoader code that might throw, skips any remaining ads (as the errors are
not recoverable, in general) and notifies a new load error callback so that the
application can implement its own handling. The intention is to make the loader
robust to unexpected requests from IMA and avoid crashes.

Also handle IMA loading an ad in an ad group that has no available ads. In rare
cases IMA will try to load an ad for which an error was previously notified, so
this drops those load requests allowing playback of the content to continue.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185985850

Issue:#3340

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185973510

All media periods are part of a queue of windows buffered and played by
ExoPlayer. When repeating windows, the current MediaPeriodId is insufficient
to distinguish between the repetitions of the same period. This makes it hard
to see to which media period load events belong to, and it is also difficult to
determine whether two media periods belong to the same logical window or
whether they are part of different repetitions of the same window.

Therefore this change adds a unique sequence number to each window in the
sequence of windows and this sequence number is saved as part of the
MediaPeriodId.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185829509

If there was no preroll and the pending content position was set before the
first midroll, the pending content position was never cleared so loading the ad
was never triggered.

Only set a pending content position if we know that we need to trigger playing
an ad for the current position and IMA will poll for an ad (because there is a
midroll ad group). Clearing the pending content position happens when IMA pauses
content to play the ad.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185818315

IMA sometimes delivers an ad group load error just after the time of the ad
group, which is problematic now that we set the expected ad group index based on
the last returned content progress.

Only update the expected ad group index once we are within a fixed preloading
threshold of the next ad.

Also fix updating the ad group to use the new ad count, and check for ad group
load errors when we have no expected ad group defensively.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185803086

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185681751

Content progress is only polled if there are midroll ad groups. If the ad tag
had only preroll/postroll ads, the pending content position was not provided to
IMA, which meant that the postroll ad could never play.

Only set the pending content position if there are midroll ads.

Issue: #3715

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185680803

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185676652

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185526653

Issue: #3782

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185673731

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185673454

Issue:#3792

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185668919

This error marks the current ad group as unplayable.

Issue: #3801

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185663472

Also added comments for all existing devices for easier reference.

Issue:#3835
Issue:#3236

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185661900

Also make ad group skipping more robust. After calling onError for an ad, IMA
will sometimes trigger an ad group load error, so this needs to be handled in a
way that allows some ads to be loaded already for the ad group.

This change also fixes calculation of the expected ad index to take into account
whether the position is being faked to trigger loading an ad or is the actual
player position.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185655844

Releasing the player released the internal playback thread once renderers were
released. Releasing a MediaPeriod queued a Loader.ReleaseTask on the loading
thread which would post back to the playback thread. If the playback thread had
been quit by the time this happened, the release task wouldn't be run.

Release on the loading thread instead of the playback thread. This avoids
needing to block releasing the player until the loading threads have ended, and
ensures that release tasks will run eventually. As part of this change,
ExtractorMediaPeriod's call to Extractor.release will now run on the loading
thread (which means that all Extractor methods are called on that thread) and
other cleanup in ReleaseCallback will run on the loading thread instead of the
playback thread.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185651320

Issue: #3724

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185548632

Issue:#3835

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185501181

IMA's cue points may not be in order, so sort them. It looks like IMA events use
time ordered ad indices, so it is not necessary to map between the original cue
point order and the time order.

Issue: #3716

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185495798

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185409304

There are 4 tests which can't currently be moved:
 - DownloadManagerTest explicitly uses the main looper which isn't easily
   supported because the test runs on this thread.
 - ContentDataSourceTest uses an AssetFileDescriptor which wraps a
   ParcelFileDescriptor. It seems Robolectric doesn't correctly forward the
   inner wrapped file descriptor leading to NPE.
 - CacheContentIndexTest and SimpleCacheSpanTest both work fine with Gradle
   but fail with seemingly valid test failures on Blaze.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185366678

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185139106

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185138158

If the period uid doesn't match, the update procedure currently doesn't
remove the correct periods. This may cause the player to get stuck or to play
the wrong periods.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=185129503