relay: isolate relay state on dedicated executor to support multiple I/O threads

[afrind]

---

• test: add MultiThread relay test mode #363
• -> relay: isolate relay state on dedicated executor to support multiple I/O threads #362
• relay_thread config and allow > 1 thread in config #361

---

This change is 

[akash-a-n]

@akash-a-n reviewed 25 files and all commit messages, and made 2 comments.
Reviewable status: 25 of 26 files reviewed, 1 unresolved discussion (waiting on afrind, gmarzot, michalhosna, mondain, Oxyd, peterchave, suhasHere, and TimEvens).

---

src/MoqxRelay.cpp line 872 at r4 (raw file):

    }
  }
  if (session.get() == upstreamSession.get()) {

is self fetch not a concern anymore? Claude just flagged this.

[michalhosna]

@michalhosna made 1 comment.
Reviewable status: 25 of 26 files reviewed, 2 unresolved discussions (waiting on afrind, gmarzot, mondain, Oxyd, peterchave, suhasHere, and TimEvens).

---

src/MoqxRelay.h line 129 at r4 (raw file):

  // cross-executor filters. relayExec must outlive this relay.
  // If not set (default), all operations run on the calling thread.
  void setRelayExec(folly::Executor* relayExec) { relayExec_ = relayExec; }

It seem unsafe to call setRelayExec after the relay starts to do anything, so why is it not in constructor? Or some kind of safe-check that it's called only once.

Looking at #363 its used there in resetRelay in the test fixture, but I am not sure it's correct. When alling this on relay where ownedRelayExec_ != null, the ownedRelayExec_ stays dangling, that's probably fine?

Is this whole method with sharing executor tests only? Should we gate it more?

Reviewing this one without #363 seems to not make much sense. I'll redo my review there.