GlobalKaleidoServer is not thread-safe

[emilykl]

Reported by @talleibman in #431:

We hit an issue in production where concurrent call_function calls occasionally returned each other's results. The shared _return_queue has no guarantee that the caller who put the task gets the matching result back.

[ayjayt]

@emilykl I think this would also be solved by 431 right? these are all the same issue.

[emilykl]

@ayjayt I believe this issue is resolved by #431, yes.

However as I just commented in #441, I believe that one is NOT resolved by #431.

[ayjayt]

Ill review again, I do remember in the meeting they'd be treated as separate issues, but I do see they share the same pathway. The hang I guess would not necessarily be a result of race conditions between tasks. This still feels like api abuse due to threading into the simplified API...