[EventSource] Block EventCommandExecuted callback during initialization

[mdh1418]

During managed startup, the managed EventSource RuntimeEventSource is initialized, kicking off EventSource initialization where all deferred commands m_deferredCommands for that EventSource instance are processed one at a time at the end of EventSource.Initialize.

However, should a derived EventSource set its callback EventCommandExecuted within its OnEventCommand, then because of the decision to invoke the callback on all deferred commands, the callback will be invoked before all deferred commands could be initially processed in EventSource.Initialize.

CounterGroup registers its callback during construction, so should multiple events be sent to RuntimeEventSource during initialization (e.g. m_deferredCommands = [ Update, Update, Update, ... ]), then because RuntimeEventSource initializes multiple event counters once the deferred command is processed at the end of EventSource.Initialize() ((i.e. m_deferredCommands = [ Enable, Update, Update, ... ]), then the callback is invoked on all deferred commands before they are converted from Update.
This leads to an assertion error as demonstrated by

EventSource[2000003047]Invoke deferred command on this value:System.Diagnostics.Tracing.EventCommandEventArgs(-1391687009 Command = Enable)
EventSource[2000003047]Invoke deferred command on this value:System.Diagnostics.Tracing.EventCommandEventArgs(663802102 Command = Update)

in #96324 (comment).

This PR aims to avoid double-invocations of the same callback m_eventCommandExecuted during EventSource.Initialize should the more-derived EventSource class set-up its EventCommandExecuted callback within its OnEventCommand.

To do so, we defer invoking the EventCommandExecuted callback on deferred commands until after all deferred commands are first-processed during initialization.

Fixes #94964
Fixes #96324
Fixes #99497
Fixes #92519

[dotnet-policy-service]

Tagging subscribers to this area: @tarekgh, @tommcdon, @pjanotti
See info in area-owners.md if you want to be subscribed.

[tommcdon]

LGTM!

[noahfalk]

This fix looks like it works around the issue which is OK, but I think we could fix the root cause fairly easily. Comments inline.

[noahfalk]

Functionally the code change looks fine but I recommended some tweaks to the comments.

[davmason]

Thanks for tracking this down!