clone() always returns a new object, which wasn’t being used in
this case. 

Some decoders will throw randomly, that doesn’t mean that we can’t 
continue on to try other decoders. 

Fixes #2406.

Fixes #2407

Group resource decoders into buckets, where the user can specify a bucket ordering so that prepending and appended is localized to the bucket.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=168779283

It's pretty expensive to do, so we're wasting time for videos. If
setDataSource is called with a non-video, it will throw an exception
which in turn will cause us to move on to the next decoder, so the
behavior shouldn't change.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=168390557

Fixes #2059

Progress towards #2354

Fixes #1510.

Fixes an issue where setting options on the generated GlideRequest after
having used it to load one image would throw an exception because the
internal RequestOptions object was locked when the first load was 
started. Now we apply autoClone() to make sure that the options object
for the first load isn’t modified by subsequent changes to options, but
also allowing those changes to occur.

Progress towards #2379.

It seems like this shouldn’t be necessary, but it fixes a PMD warning.

Fixes #2394.

The annotation can only be added to final methods, so we have to use
@SuppessWarnings for RequestOptions instead. As a result callers of
transforms() will have to suppress the warning on their end as well.

Fixes #2383.

Also fixes an issue where if we failed to find a support Fragment in
a FragmentActivity, we’d never check for a non-support Fragment in the
same Activity.

Fixes #2370.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=168236016

floor is necessary to make sure that we populate all pixels in the
Bitmap, but it's not necessary to use to check if our input Bitmap has
already been transformed and can be returned as is. Using floor can
cause us to re-draw a Bitmap unnecessarily.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=167789795

This allows models to handle equity for internal states that can be different from the target image result that is handled with equals()/hashCode()
For example to allow restart of image download from a new source but still using the same cache in Glide.

In addition to v3.0.1 being super old, it looks like earlier version(s)
of OkHttp have a race that will leak resources when requests are
enqueued instead of executed and cancel is called.

Progress towards #1996.

Fixes #1063.

Works around a ClassCastException in OkHttp on O due to a framework bug.

Fixes #2355

Previously some of the logic in run() might null out the data fetcher
instance variable, even if it had been set when run() was first called.
As a result, the data fetcher would not be cleared. Now the initial 
data fetcher is held on to as a local variable so the instance variable
can be freed without affecting the behavior of cleanup().

Progress towards #1996, #2352

Fixes #2331.

The Key we’re given may have a different implementation for equals and hashCode than it does for updateDiskCacheKey. As a result, two different
Key objects that are not equal to each other may produce the same
disk cache key. To avoid simultaneous puts for the same disk cache key
we need to lock on the disk cache key, not the original key object.

* Add prepend methods for EncoderRegistry and ResourceEncoderRegistry classes

That way we will always close the stream in cleanup(). Prior to this change, when cancel was called, stream was never set to non-null and therefore never closed in cleanup. Because disconnecting the stream doesn’t seem to close the contained ResponseBody object, failing to obtain and then close the stream results in leaked objects.

Progress towards #2352

Progress towards #2352.