Remove all ElementUtil allocations and stop boxing in DispatcherOperation

[h3xds1nz]

Fixes #10679, currently blocked by #10680.

Description

This is a bigger PR, however the impact is proportional to the size of it.

• Removes all allocations in ElementUtil of DispatcherOperationCallback.
• Removes all various boxing during invocations and on returns in all automation wrappers / ElementProxy.
• Removes boxing of TResult when the invocation is not on the same thread / Send priority in DispatcherOperation.
  • It didn't box on same thread / Send priority but the rest was using the object Result.
• Adds some basic invocation tests for Dispatcher/ElementUtil.
• Adds nullability annotations to all UIA proxy wrappers (and some interfaces) to further support the effort.
• Adds basic unit tests for ValueProviderWrapper / ToggleProviderWrapper / RangeValueProviderWrapper / ExpandCollapseProviderWrapper / TransformProviderWrapper that test for both wrapper construction and invocations.

To further reduce the code complexity between legacy/async invocations, the different flavours of DispatcherOperation only contain the respective code-paths that are required, e.g. the non-legacy paths no longer contain wrapped invokes that would never be invoked (as _useAsyncSemantics would be true).

• Making DispatcherOperation abstract and removing the protected virtual member is not a breaking change as there's no public/protected constructor and thus nobody could have subclassed it or used the method call via regular means.
• See https://github.com/dotnet/runtime/blob/main/docs/coding-guidelines/breaking-change-rules.md#types

Overall, this will now cause inlining on all the regular hot paths in ElementProxy in tier1, further speeding up most of the synchronous/same-thread invocations (which is often the case).

There's one thing to consider:

• The only way the ElementUtil inner operation won't complete (when the result would have HasCompleted set to false) is when a CancellationToken would be used - which is never, making this essentially a dead code. But I'm keeping it as-is.

Customer Impact

See #10679 for impact of this change. In the demo app (which is a repro from this repo), it reduces the total allocation in half and reduces the time required (big part is due to allocations, some due to inlining etc.) to complete by 30% on average.

The unit tests here will also help verify changes in #10237.

Regression

No.

Testing

Local build, new unit tests, testing with regular apps and UIA invocations.

Risk

Medium, the changes are quite broad yet simple. I've included numerous tests to ensure functionality.

Microsoft Reviewers: Open in CodeFlow

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 1293 lines in your changes missing coverage. Please review.

Project coverage is 2.99275%. Comparing base (57fe8ac) to head (3935c32).
Report is 14 commits behind head on main.

Additional details and impacted files

@@ Coverage Diff @@ ## main #10903 +/- ## =================================================== - Coverage 13.48885% 2.99275% -10.49610%  =================================================== Files 3317 3338 +21 Lines 664749 664807 +58 Branches 74643 74729 +86 =================================================== - Hits 89667 19896 -69771  - Misses 572488 644726 +72238  + Partials 2594 185 -2409 

Flag	Coverage Δ
Debug	2.99275% <0.00000%> (-10.49610%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.