Fix race condition on simultaneous hot restarts

[annagrin]

Make hot restarts atomic

• Replace uses of Pool(1) by an AtomicQueue wrapper for clarity.
• Execute hot restart service extension atomically.
• Add test for simultaneous hot restarts.

Towards: #1869

[elliette]

Could we break this up a bit more?

PR 1, addressing #1869:

Fix race condition on running two createIsolate calls simultaneously:

• Run destroyIsolate followed by createIsolate an atomic operation.
• Return error on expression evaluation if the evaluator is closed.

PR 2, addressing #1704:

Make debugger API that depend on isolate running wait for the start
of the app.
Note: I think in that PR we should use the observer pattern as mentioned in #1870 (comment), as well as re-enable the tests that were disabled in #1751.

PR 3, cleanup:

Remove unnecessary async keywords from synchronous debugger API

[annagrin]

Current PR breakup plan:

1. Synchronize hot restart (this change)
2. Make expression compiler reject requests if closed
3. Update debugger and vm_service state machine to make sure they are in sync with each other
4. Remove unnecessary async keywords from the debugger and re-enable getStack test

[elliette]

LGTM with one request - can we rename runTask to simply run (since from my understanding it will run all tasks in queue sequentially?)