Debugger launch has a race condition making breakpoints in startup code are unreliable

[gregg-miskelly]

1. Create a new process suspended
2. Use RegisterForRuntimeStartup to find out about the new process
3. When RegisterForRuntimeStartup fires the handler, call DebugActiveProcess and return

Expected:
Debugger should successfully finish the hookup before any managed code runs

Actual:
There is a race where sometimes managed code runs before the hookup finishes. CORDebuggerAttached gets set from coreclr!Debugger::MarkDebuggerAttachedInternal, which is being called from coreclr!Debugger::SuspendComplete.

We have seen this on various Linux distros. I can't tell for sure if this is an xplat-only problem (or maybe Linux only?) or if this applies to Windows also.

[mikem8361]

Fixed in master in PR: dotnet/coreclr#8951

[mikem8361]

1.1.x PR dotnet/coreclr#9060, 1.0.x PR dotnet/coreclr#9062