Faster joins: improve handling of race where we can fail to persist a client event with partial state after _sync_partial_state_room clears the partial state flag for a room

[squahtx]

Follow on from #12988.

Currently, when we encounter the race in #12988, we send a 429 Too Many Requests back to clients with the expectation that a retry happens.

@reivilibre notes that some clients may require retries to be manually initiated, which is not good UX.

We can improve things by doing some of the retries internally in synapse.

---

A handy plot of code paths which create client events:

The 429 gets raised in EventCreationHandler.handle_new_client_event. Event contexts are computed in orange methods.

[squahtx]

@reivilibre spotted an occurrence of the race during a complement run: https://github.com/matrix-org/synapse/runs/7397381295?check_suite_focus=true#step:4:2049
This one looks like it's on the _local_membership_update path

[reivilibre]

If we want to just make Complement retry the request, this will do it: matrix-org/complement#414

I don't know if that's something we actually want to do or not?