SocketOutput's Monitor.TryEnter(_contextLock) burns cpu

[benaadams]

The fast fail Monitor.TryEnter in SocketOutput introduced by #562 is more expensive than it first seems when contended.

I originally saw high cpu burn in locks in Kestrel in this issue https://github.com/dotnet/coreclr/issues/6144 but wasn't sure if it was just the normal locking; looks like it can also be caused by extra contention from the TryEnter atm 😞

See (or blocked by) https://github.com/dotnet/coreclr/issues/6950

[benaadams]

Is fixed in coreclr now, close?

[halter73]

@benaadams Thanks for the fix 😄