Remove RequiresPreviewFeatures attribute on Lock, add test using the lock keyword

[kouvel]

• Also removed the EnablePreviewFeatures project properties that were added with Lock changes
• Added some basic tests with Lock using the lock keyword

[ghost]

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

[dotnet-policy-service]

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

[stephentoub]

Are we ready to start using Lock in runtime? Would the general philosophy be to use it anywhere we're currently using a dedicated lock object and it's only being used for enter/exit and not signaling?

[mangod9]

yeah we were planning to initially start using it in Threadpool synchronization logic. But might be good to start using in few other places too, so we can iron out issues (if any) before going broader.

[kouvel]

Yea we were planning on using it in a few places relevant to the thread pool to start with. Those locks are used frequently but aren't very contentious. It is already used in NativeAOT, wanted to get some more coverage in CoreCLR as well. It should be ready to be used anywhere Monitor is used currently just as a lock. There are some SOS diagnostics to add, but it should be fine to update existing uses.

[stephentoub]

Perf wise, how should I think about it? It'll be as-good-or-better in pretty much any scenario, such that I can use it without doing significant perf validation, or it's more nuanced?

[kouvel]

Lock perf was >= perf of Monitor in pretty much any case I tried (with EnterScope or lock keyword). It is a bit nuanced if Enter/Exit are used instead, as the thread-local access is more expensive and there are other code gen issues that can make it a bit slower in some cases. Even with Enter/Exit, it really depends on what is done inside the lock and how good the processor is at prefetching, and in most cases the perf was still equal or better.