stub: try enable GRPC_CLIENT_CALL_REJECT_RUNNABLE

[YifeiZhuang]

Did dryrun tests on cl/435778399, enabling the flag, for multiple times, not seeing failures cases that are obviously caused by our change, so I am willing to give it a try. @ejona86 Wdyt?

Tests tended to be flaky. But the latest TAP train test signals that google3 might be ok if we enable it:

• latest TAP results, initially some are broken
• rerun failure successful

[ejona86]

The tests look good. Although we still need to fix the cases in ClientCalls.blockingUnaryCall() where we shutdown the executor without waiting for the RPC closure.

[YifeiZhuang]

The tests look good. Although we still need to fix the cases in ClientCalls.blockingUnaryCall() where we shutdown the executor without waiting for the RPC closure.

you mean the incremental fix that we only shutdown the executor when RPC is successful (status is ok)?
we were talking about failure cases that we shutdown RPC without knowing if they are closed.
On those failures cases I assume we should not keep executing runnables that is added to ThreadlessExecutor.execute() ?
I think we can merge this change first? It is more strict so more risk, but would be easier to rollback.

[YifeiZhuang]

we discussed this. Only enable when ThreadlessExecutor.shutdown guarantees that is happens after call is closed. shutdown is changed in #9042

[ejona86]

Assuming #9041 or #9042 goes in before this, this LGTM