Move to xunit 2.3.1

The new analyzer warnings are valid but are quite noisy. Going to
suppress them for now to unblock this upgrade. Will fix them in a
separate change.

* moved waiter from diagnostics.dll to features.dll where all interfaces are defined.

* simplified listener and waiter and moved those in product.

* cleaned up a bit

* fixed some comments

* forgot to add new AsyncListenerProvider in test export composition

* fixed dead lock found on unit tests

* remove all waiters left

* added dispatcher event processing in one more place

* added a way to enable waiter through env variable in test such as DDRIT

* added env var support so that It can be used without internalvisibleTo us.

* dont wait for all waiters but only onces each test used to wait.

* removed dependency to Roslyn.Services.Test.Utilities in integration test. that dll is not in probing path.

* fixed test failures

* fixed bug in integration tests

* move call to enable waiter before creating workspace so that workspace event queue get proper async listener

* expose concret type as well so that it can be directly used in test for waiter

* addressed PR feedbacks.

* rename mess up

* extra ) left out

* fixed build failure

* Simplify diagnostic tagging by making it use the standard tagging model.

* Cleanup.

Make analyzer diagnostics independent of the current warning level

Fixes https://devdiv.visualstudio.com/DevDiv/_workitems?id=548762

Block use of  explicit `in` arguments in dynamically dispatched expressions.

[Perf] Pre-allocate array builder sizes for loading PE members

The LDM has decided that the following rules should be in effect when
calculating definite assignment for local functions:

  1. The entry point to a local function is always reachable.
  2. Variables captured in local functions are definitely assigned if
  they are definitely assigned in all branches into the local function.

It turns out these rules were almost exactly what the compiler already
implemented, but there was a bug in captured variable detection that
meant that variables captured in lambdas within local functions were
sometimes not counted as captured. This change fixes the bug around
capturing, which should cause the compiler to conform to this
specification.

Fixes #17829

Explicitly marking an argument as an `in` argument has effect on overload resolution and needs to be known to the dynamic binder to be able to make appropriate decisions. For the time being there is not way to pass this additional information.
That would require a small expansion of APIs used by `dynamic`. When we have such API, we should predicate this scenario on the presence of such API.

For now, In order to not perpetuate accidental and likely incorrect behavior we should block this scenario.

Fixes:#22813

The argument that stands for the receiver of a `ref` extension method must be always passed as `ref`

* Make sure 'in' is honored in cref parsers/completion lists

* Clean up

* PR Comments

* Document API breaking change

This looks like simple oversight. The binder being used for type
parameters was retrieved from the binder factory since we didn't
explicitly pass one. For members this is correct since the binder
factory can be queried for top-level binders. This is not correct for
local functions because they use method body binders, which are not
accessible from the binder factory.

Mostly, this doesn't matter. The exception is when you try to
speculatively bind an attribute on a type parameter. Here, you need an
in-method binder and VS will crash if it's the wrong binder. This was
uncommon since attributes on type parameters are not permitted in local
functions, but VS should not crash.

Fixes #17814 

Merging on behalf of @zaytsev-victor. Thanks for the contribution!

* Arrays/pointers/nullables of intrinsic types are intrinsic. Type of array creations is apparent. Handle out vars.

The argument that stands for the receiver of a `ref` extension method must be always passed as `ref`

Snap master into dev15.6.x