Reject struct promotion of parameters when -GS checks are enabled

[briansull]

This causes bad codegen when we introduce shadow copies for them.

[briansull]

Introduced with #16220

[briansull]

@AndyAyersMS PTAL
@dotnet/jit-contrib PTAL

[briansull]

I will try to work up a repro case for this issue.

[AndyAyersMS]

Can you verify we still get the same codegen for the span-stackalloc cases in #16197?

Seeing running diffs would be valuable. Likely there should be no diffs in core and framework assemblies. Perhaps a few in tests if we have some IL tests that have unsafe value classes.

[briansull]

I saw a handful of diffs on the desktop. '
It loosk like IL_STUB's sometimes have an unsafe buffer.
The diffs I saw were typically just in the locals variable names used in the header.
Often the generated code was exactly the same.

[briansull]

One such diff occurs in
; Assembly listing for method System.Reflection.Associates:AssignAssociates(struct,int,ref,ref,byref,byref,byref,byref,byref,byref,byref,byref)

where local 4 is an unsafe-buffer
; V15 loc3 [V15 ] ( 8, 11.24) struct (80) [rsp+0x20] do-not-enreg[XSF] must-init addr-exposed ld-addr-op ptr unsafe-buffer

That struct is a MetadataEnumResult

which is defined:

internal unsafe struct MetadataEnumResult

So it looks like there are two ways to mark a struct as being an unsafe-buffer

Either use the full Attribute name like I did in my test case:
[UnsafeValueTypeAttribute]
struct DangerousBuffer

or by simply using the C# keyword unsafe in your struct definition as is done for MetadataEnumResult:

internal unsafe struct MetadataEnumResult

[AndyAyersMS]

Thanks for fixing this.