It seems there's a flaky issue in our unit test setup where sometimes
ThrowIfNotOnUIThread will throw even if you're on the JTF main thread.
It appears the implementation relies on checking the Dispatcher object
that must sometimes not be set up right. We have a standard mechansim
anyways for asserting that we are on the UI thread that consumes our
IThreadingContext, so switching the helper we call should avoid this.

Backport editor config change to 16.4

Ensure that we show the gold bar for .editorconfig as solution item o…

Core changes to prep for inferred indentation

* Change inline rename dialog to be more uniquely identifiable by automation id

* Add an important note about the automation id needing to remain

We never hooked up the options provider, which meant syntax trees
would have the options but not the documents themselves if you asked
for an OptionSet.

This exposed an issue in our .editorconfig configuration fix tests:
those tests were setting dotnet_style_object_initializer = false and
then trying to assert that the configuration fix comes up. This setting
should have had the effect that the diagnostic was being disabled
and no fix would be triggered. But since .editorconfig was being
ignored in calls to GetOptions() the fix triggered anyways and
everything else worked from there. Once we fix the .editorconfig bug,
the tests start failing because there are no longer fixes. I'm updating
the tests that change the severity to simply pass true since they
weren't actually sensitive to what the boolean was. The tests that are
trying to test setting a value to true will need to find a different
option entirely to test with, and that's tracked by
https://github.com/dotnet/roslyn/issues/39466.

There were some tests that generally tested the behavior that "if you
close a document, the tracking span should go away". These tests weren't
actually working right:

1. The AssertMissingPastedTextSpan helper fetched the text container
   and passed that to the PasteTrackingService. If the documents had
   been closed, then the container returned is a throwaway container
   which isn't a text buffer at all. The PasteTrackingService returns
   no span, but it always would have no matter what. As such, this case
   meant this was less a test of the service and more a test of
   TestWorkspace that the buffer was closed when expected.
2. If the documents were open at time, a buffer might get reused by
   TestWorkspace. In those cases, the product code also has no code
   that was clearing the buffer properties out: the test code did.
   As such, those tests passing weren't asserting product behavior,
   they were just asserting the test harness behavior.

In practice, any attempt to resurrect these were really just tests about
the test harness or TestWorkspace, and not product code. So deleting
them is just the right thing.

The only remaining uses now are ones actually testing the core workspace.

We had both a property and a method returning the same thing. This
unifies to just the method.

With TestWorkspace we were creating text buffers for all types of
documents up front, and by default opening all buffers in the workspace.
This was problematic because if we're creating text buffers we should
be ensuring the text buffer's options for using tabs and spaces matches
the Workspace options, but the workspace sometimes wasn't available
at those times. This is also unnecessarily expensive if you have test
projects that aren't doing anything with editor experiences at all.

Also, a complicated "create a separate document in the TestWorkspace for
the cursor" was removed in favor of more direct code that just creates
a buffer. Creating an entire TestDocument was funky since the document
was never actually part of the workspace -- this revealed that it didn't
seem we were ever disposing the views created.

The "languageName" parameter to CreateProjectionBuffer was removed: it
didn't logically make sense (the projection buffer created isn't in
a Workspace, so it has no language name), but also wasn't being used
for anything else.

* Add solution crawler wrapper types

* Add Experimentation wrapper

* Add UnitTestingPinnedSolutionInfoWrapper

* Add KeepAliveSessionWrapper

* Add PersistentStorageLocationService wrapper

* Add UnitTestingWorkspaceTaskSchedule wrapper

* Add extension wrapper

* Change Workspace wrapper to an extension class

* Add more wrapper types

* Rename and remove files

* Add immutable array extensions

* remove lut ivt

* Remove the dependency on ServicehubBase

* Remove all the dependency for LUT

* Rename and remove unused wrappers

* Add comments

* Rename wrappers and accessors

* More renaming

Co-Authored-By: NSukru Alatas <alatas@gmail.com>

Fixes #39472

Adjust syntax equivalence to account for nullable directives

Show code instead of designer, but only after trying to create the designer loader

The overloads sometimes had exportProvider first and sometimes second.
This was because if you always had it first and content type is passed
as a string, you have ambiguity of how to handle the params array.
Just passing in stuff explicitly makes this easier.

Also replace a hand-written implementation of string.Join with
string.Join.

Makes it easier to figure out when just one side is failing.

This is implemented by adding metadata to exports of
IDocumentOptionsProviderFactories that specify the ordering like any
other provider. Although I would love to simplify this by simply
removing the factories entirely by simply having the Workspace
constructor directly add the instances, Visual Studio for Mac is also
exporting these so we'll maintain compat for now.

I expect that IOrderableMetadata existed back from a time where we
were working with both MEFv1 interface metadata views and MEFv2
concrete type views. Nothing today needs it anymore, so we can just
delete it.

If we both had a document-specific option provided and also tried
to change the value, we'd end up throwing an exception.

This adds a convenience overload where WithChangedOption can be more
easily used with a PerLanguageOption, but also returns a strongly
typed DocumentOptionSet so you don't lose the distinction.

These no longer apply.

There appeared to be one bug: if a document was opened that had a null
file path, there was a risk that we would crash because we would be
passing that around to the editor API in the legacy case.

Handle intellicode star in display text prefix

Fix IDE test

Change ignoreChildNode to fast pass on AreNullableDirectivesEquivalent. Fix comparison of imports to ignore nullable trivia

Semantic model assert

Link to http://sourceroslyn.io

Performance fix in analyzer driver for CFG based analyzers

Re-use semantic model when attempting to compute suppress message att…