The test caching layer had two bugs around re-hydrating the test resutls from cache:

- Made a bad assumption about the name of the file.
- Assumed tests were always rehydrated into the original directory.

closes #8860

Revert "Revert "Merge pull request #7510 from CyrusNajmabadi/analysisSpeedup""

Revert "Merge pull request #7510 from CyrusNajmabadi/analysisSpeedup"

This reverts commit 4ed5b596, reversing
changes made to 29ab8cba.

Ask Mode request: Fix IndexOutOfRangeException when running operation actions over Roslyn sources.

Fix vbc crashing when compiling 0 files into a library 

Ignore dlls that come from a 'packages' directory when searching for viable metadata references to search.

Fixes #8577

Clarify behavior when RunServerCompilation throws an exception

Reuse the SupersedeLowerVersions flag in the EE

Move interactive commands out of InteractiveFeatures

Disable Failing Documentation Tests (Stabilization)

Scripting scenarios set a special flag in the ReferenceManager to enable
additional tie-breaking rules in the presence of conflicting references.
Enable that same flag for the EE, which also routinely encounters
conflicting references.

Fixes internal bug 161294.

Match equivalent symbols in non-identical referenced assemblies

Handle VB property with type character when calculating trivia changes during EnC

Ignore dlls that come from a 'packages' directory when searching for viable metadata references to search.

A question was raised about the behavior of the compiler when RunServerCompilation throws an exception. After looking through the code I decided the approach most consistent with our existing code is:

1. Stop using `null` as a valid value in this API.  If the client determines it can't make the call use the Rejected coede.
2. Gracefully handle case where Task is faulted.  This puts the IO handling in a single place.
3. If the compilation could not complete on the server, for whatever reason, always fall back to local compilation.

closes #7866

Merge from master

made roslyn to use new GetContentLocation API from extension manager

added column filter to suppression column so that it doesnt count in …

turn off low latency mode when we hit low memory situation.

Merge pull request #8655 from jasonmalinowski/remove-silly-resolveassemblywarnorerrorOntargetarchitecturemismatch

Remove useless ResolveAssemblyWarnOrErrorOnTargetArchitectureMismatch…

This appears to have exposed a bug in the compiler.  Disable checking for now.

see #8739 for more details.

IOperation conditional access expression API change

Renamed more VB private members that implement interface members using the convention of InterfaceName_MemberName.

the low latency mode is introduced in the early days of Roslyn to make sure we always hit 50ms typing responsive time in our own perf lab typing test.

basically that mode prevent GC to get into most expensvie blocking gen 2 collection. which is nice most of time, but when we are under low memory sitatuion, not doing blocking gen 2 also means we don't do compact.

and that I believe can cause us to be in a sitatuion such as https://github.com/dotnet/roslyn/issues/6802

this basically let us turn that optimization off sacrifying keystroke responsiveness time to time to have more memory available to us.

this mode is supposed to force blocking gen 2 collection when GC detects its own low memory condition, but that seems not consider when there is some free memory left but all those are fregmented.