Multiprocess Stage POC

[cwharris]

Example demonstrates the use of Python Multiprocessing with a custom Stage, which enables offloading of MultiMessage processing to a separate process while continuing the workflow within the same process.

The example results in a segfault originating from Watchable::watcher_epilogue when calling sub.on_next from python in the parent process. It appears that one of m_watchers has been released and we're still trying to access it. Unfortunately I don't know enough about MRC to say what the cause is.

inline void Watchable::watcher_epilogue(WatchableEvent op, bool rc, const void* addr)
{
    for (const auto& obs : m_watchers)
    {
        obs->on_exit(op, rc, addr); // <- segfault occurs here, likely when de-referencing `obs`
    }
}

I was able to work around the issue by simply commenting out the obs->on_exit(...) call and rebuilding MRC and Morpheus together. Everything seemed to work fine, so maybe this is a bug in MRC?

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[jarmak-nv]

Hey @cwharris should we promote this out of draft since you're requested review?