MSC4140 delayed event rate limit failures are not handled

[hughns]

As per https://github.com/matrix-org/matrix-spec-proposals/blob/toger5/expiring-events-keep-alive/proposals/4140-delayed-events-futures.md#rate-limiting-at-the-point-of-sending we are applying rate limiting at the point of the delayed event being entered into the DAG.

However, we don't handle this temporary failure and retry.

This means that the delayed event then gets dropped.

This was observed in the wild yesterday.

[hughns]

#18018 contains a proposed fix.

However, the approach in that PR is to simply disable that rate limit.

If that PR is accepted then the should be some follow-up on this issue here to get the original security consideration mitigated. For example, one of:

• taking account of delayed event scheduling when accepting the delayed event in the first place
• adding the rate limit back in but handling the rate limit failure sensibly