Do not reset size of TailCallArgBuffer

[jkotas]

TailCallArgBuffer:State is 32-bit int. The code incorrectly used native int operation to set. It lead to TailCallArgBuffer::Size to be reset as well and needlessly re-allocating the tail call buffer.

Fixes #118166

[Copilot]

Pull Request Overview

This PR fixes a bug in the tail call buffer implementation where the State field was being incorrectly set using a native integer operation instead of a 32-bit integer operation, causing unintended resets of the Size field and unnecessary buffer re-allocations.

Key changes:

• Changes EmitSTIND_I() to EmitSTIND_I4() when setting the TailCallArgBuffer state to ensure only the 32-bit State field is modified

[dotnet-policy-service]

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

[jkotas]

This should fix the stress problem hit by the CI. I have noticed that the thread in the bad state is always in the CFG probe called from free method in Debug CRT.

There is still a corner case stress bug hidden somewhere. It is likely specific to combination of WinServer 2016 w/ CFG (and potentially Debug CRT). I suspect that it may be in special casing of CFG inside SetThreadContext in the OS. It should be rare enough that we can live with it.

[jkotas]

I have realized that we can call malloc/free in pre-emptive mode to make this even more robust, going to push an update in a bit.

[jkotas]

This is ready for review again