* fixed a race where solution cralwer used wrong solution for it to process.

race is this.

1. solution crawler picked up a solution
2. before processing the solution, workitem got changed
3. and then picked up an work item from the queue
4. and use the work item with the solution that got picked up in step 1

step 2 is happening beacuse solution has changed, but step 4 used old solution from step 1
that doesn't have effects of the solution changes.

solution crawler must pick up solution after it has picked up work item from the queue so that new changes of a solution get enqueued as new work item,

* Update src/Features/Core/Portable/SolutionCrawler/WorkCoordinator.NormalPriorityProcessor.cs

thank you
Co-Authored-By: Nheejaechang <hechang@microsoft.com>

* addressing feedback

Create TempPE compiler for the new project system

Delete unused packages

Typos

Fix up the implementation of IDocumentTrackingService

Fix goto-def and glyph for builtin operators.

We get these from them now, we don't make them ourselves.

This is still needed for running unit tests, but just in one place. We
can remove it everywhere else.

Don't have 'Keyword' map to 'ReservedKeyword' and 'ContextualKeyword' it just ends up confusing people with terminology.

The previous implementation assumed that it'd get a call from the
old DocumentProvider to know when new files were opened so it could
add it to the list of frames. When I deleted the DocumentProvider, that
hookup wasn't working anymore, and so it would never be tracking
anything at all. Oops.

Rather than couple some types together, the implementation is now
more standalone. I've elected, rather than also listening to RDT events
to simply start tracking a frame the first time it's activated.
The only way I can think a document could be made visible without being
activated that we would care about is on first time load when we
re-open existing documents. If you had a bunch of documents open, and
they were in different frames/windows, we won't consider the non-active
documents to be visible until you first focus on them. I'm not sure this
is really much of a loss; we've often seen that many customers have
files reopened on them that they didn't care about, so this might
actually improve solution open performance since we won't be processing
files they may not care about.

This also brings the service up to a slightly more modern/correct
implementation. This is now a proper MEF part that doesn't assume it's
being created on the UI thread, and it manages it's own shutdown rather
than assuming the package will shut it down in the package unload.
The previous implementation also was a bit strange in that it was a
workspace service, but still had a singleton underlying the whole thing.
The singleton was broken in that different files might be in different
workspaces, but it would always give DocumentIds out for all of the
workspaces, despite it being documented as "only returns for the given
workspace". There's still a singleton doing all the Visual Studio
eventing, but there's now per-workspace logic to correctly filter it.

Unlike the previous implementation, this doesn't really do any cleanup
anymore other than ensure that the async tasks started are joined to
avoid things running after MEF shutdown. Practically, us unsubcribing
events brought no value to anything other than spending extra time
during shutdown, so we'll just leave that as is. It feels "wrong" but
it's the way many are arguing this should be done these days.

Lastly, this now uses async service provider to fetch services. Calls
for data do not block on the service provider having done that work,
since doing so would violate other parts of Roslyn assuming that we
wouldn't block on the UI. I think it's fine, since this service is
very much a tool for optimization only. It still uses JTF internally
on calls that we know are coming in from VS (namely, the .Dispose()) so
that's fine.

Initial step to separate sync namespace refactoring and change namespace service

verify that optprof scripts are not broken

update training branch

Merge dev16.0.x to master

 Adjust the way nullable annotations are represented in metadata.

Closes #30075.
Closes #30065.
Closes #29594.
Closes #29683
Closes #29662

Update metadata references in response to file changes

Merge dev16.0.x to master

Reset master versions to preview2

This ensures that if the file is still locked when we receive an event,
we won't hit an IO error and fail to read it at all. This is the same
behavior we did with the older project system interface.