This removes the hard-coded binding redirects that exist within our
source tree in favor of redirects that are computed automatically. This
takes advantage of the built-in binding redirect generation support
in MSBuild. In order to get the redirects to merge properly with
the static stuff we have, I renamed the .exe.config files to App.config
which is the name the tooling looks for.

Addresses #7612.

This reverts commit 76b3818d.

This reverts commit 2ba614ea.

This reverts commit c6a1c2c8.

This reverts commit ef1f1a43.

We attempt to bind lambda args against every applicable delegate
type that could be a matching parameter, and then select the
"best" for error recovery.
This also has the effect of removing unbound lambdas from the
bound trees, replacing them with the "best" binding for error
recovery. This makes the semantic model insensitive to any further
trial bindings (i.e. mutation of the unbound lambda cache).
Fixes #12063, #11979, #11901

Remove vestigal "extensionMethodsOfSameViabilityAreAvailable"
thereby reducing coupling between internal compiler APIs.
See also #7740 #5128, where the vestigal APIs were introduced.

Our test utilities had a number of files that were duplicated in test utility projects.  Changed them to be linked files so we didn't have duplicate sources.

Fixes #12113

* Test ExportedType table content

* Refactoring of code emitting exported types

* Remove MetadataWriter.GetExportedTypeFlags

While working on another change I noticed that our testing support for experimental features didn't actually verify the underlying feature was experimental.  Hence helpers like ParseExperimental(...) would continue to function without error when tuples / ref returns were attached to the /langver switch.  This creates maintenance issues in our test framework as it's unclear what we're actually testing.

I refactored the code involved to verify the features are actually experimental and removed some troublesome helpers.

Support for using `var` in a local function return type was already
removed in PR #7916. This PR adds a test to verify that.

Closes #10392.

Updating the compiler binding redirects to use a version range of `0.0.0.0-2.0.0.0` to match the IDE behavior.

The test fired cancellation then waited 100ms for cancellation to be acknowledged.
There's no guarantee on a heavily loaded machine that cancellation will be
acknowledged in that time frame. This test should be reliable, so this 
removes the time out to prevent flakiness. Fixes #11819