Reduce ExecutionContext register handling code size

[benaadams]

Brought up in #21875 (comment)

Remove executing on the ExecutionContext with error handling out of flow to RunReturningAnyException methods; away from setting and restoring ExecutionContext to reduce the need for manual enregistering workarounds.

Total bytes of diff: -1955 (-0.05% of base)
    diff is an improvement.

Total byte diff includes 2397 bytes from reconciling methods
        Base had    0 unique methods,        0 unique bytes
        Diff had    4 unique methods,     2397 unique bytes

Top file improvements by size (bytes):
       -1955 : System.Private.CoreLib.dasm (-0.05% of base)

1 total files with size differences (1 improved, 0 regressed), 0 unchanged.

Top method regessions by size (bytes):
        2074 : System.Private.CoreLib.dasm - AsyncMethodBuilderCore:RunReturningAnyException(byref):ref (0/25 methods)
         174 : System.Private.CoreLib.dasm - ExecutionContext:RunReturningAnyException(ref,byref):ref (0/2 methods)
          84 : System.Private.CoreLib.dasm - ExecutionContext:RunReturningAnyException(ref,ref):ref (0/1 methods)
          65 : System.Private.CoreLib.dasm - ExecutionContext:RestoreChangedContextToThread(ref,ref,ref) (0/1 methods)

Top method improvements by size (bytes):
       -3974 : System.Private.CoreLib.dasm - AsyncMethodBuilderCore:Start(byref) (25 methods)
        -168 : System.Private.CoreLib.dasm - ExecutionContext:RunInternal(ref,ref,byref) (2 methods)
        -129 : System.Private.CoreLib.dasm - ExecutionContext:RunInternal(ref,ref,ref)
         -81 : System.Private.CoreLib.dasm - ExecutionContext:RunFromThreadPoolDispatchLoop(ref,ref,ref,ref)

8 total methods with size differences (4 improved, 4 regressed), 17548 unchanged.

https://github.com/dotnet/coreclr/pull/21908/files?w=1

/cc @stephentoub @jkotas not sure if this is a help or hindrance?

[benaadams]

Total bytes of diff: -1012 (-0.03% of base)
    diff is an improvement.

Total byte diff includes 2332 bytes from reconciling methods
        Base had    0 unique methods,        0 unique bytes
        Diff had    3 unique methods,     2332 unique bytes

Top file improvements by size (bytes):
       -1012 : System.Private.CoreLib.dasm (-0.03% of base)

1 total files with size differences (1 improved, 0 regressed), 0 unchanged.

Top method regessions by size (bytes):
        2074 : System.Private.CoreLib.dasm - AsyncMethodBuilderCore:RunReturningAnyException(byref):ref (0/25 methods)
         174 : System.Private.CoreLib.dasm - ExecutionContext:RunReturningAnyException(ref,byref):ref (0/2 methods)
          84 : System.Private.CoreLib.dasm - ExecutionContext:RunReturningAnyException(ref,ref):ref (0/1 methods)

Top method improvements by size (bytes):
       -3074 : System.Private.CoreLib.dasm - AsyncMethodBuilderCore:Start(byref) (25 methods)
         -96 : System.Private.CoreLib.dasm - ExecutionContext:RunInternal(ref,ref,byref) (2 methods)
         -93 : System.Private.CoreLib.dasm - ExecutionContext:RunInternal(ref,ref,ref)
         -81 : System.Private.CoreLib.dasm - ExecutionContext:RunFromThreadPoolDispatchLoop(ref,ref,ref,ref)

7 total methods with size differences (4 improved, 3 regressed), 17549 unchanged.

[benaadams]

A little more with next commit

Total bytes of diff: -1955 (-0.05% of base)
    diff is an improvement.

Total byte diff includes 2397 bytes from reconciling methods
        Base had    0 unique methods,        0 unique bytes
        Diff had    4 unique methods,     2397 unique bytes

Top file improvements by size (bytes):
       -1955 : System.Private.CoreLib.dasm (-0.05% of base)

1 total files with size differences (1 improved, 0 regressed), 0 unchanged.

Top method regessions by size (bytes):
        2074 : System.Private.CoreLib.dasm - AsyncMethodBuilderCore:RunReturningAnyException(byref):ref (0/25 methods)
         174 : System.Private.CoreLib.dasm - ExecutionContext:RunReturningAnyException(ref,byref):ref (0/2 methods)
          84 : System.Private.CoreLib.dasm - ExecutionContext:RunReturningAnyException(ref,ref):ref (0/1 methods)
          65 : System.Private.CoreLib.dasm - ExecutionContext:RestoreChangedContextToThread(ref,ref,ref) (0/1 methods)

Top method improvements by size (bytes):
       -3974 : System.Private.CoreLib.dasm - AsyncMethodBuilderCore:Start(byref) (25 methods)
        -168 : System.Private.CoreLib.dasm - ExecutionContext:RunInternal(ref,ref,byref) (2 methods)
        -129 : System.Private.CoreLib.dasm - ExecutionContext:RunInternal(ref,ref,ref)
         -81 : System.Private.CoreLib.dasm - ExecutionContext:RunFromThreadPoolDispatchLoop(ref,ref,ref,ref)

8 total methods with size differences (4 improved, 4 regressed), 17548 unchanged.

[jkotas]

This part (moving a part of the less common logic into non-generic ExecutionContext.RestoreChangedContextToThread) looks good to me.

[benaadams]

Let me do that as by itself as a fresh PR

[benaadams]

#21909

[jkotas]

This extra generic method is questionable because of the extra runtime metadata that comes with every generic method.

I think there is a good idea here to just have a small try/catch instead of finally. What would the gains look like if the same small try/catch is inline in the caller?

[benaadams]

Will revisit post #21909