Many copies of a large snapshot at once can be asked to retry

A copy of a snapshot that saved memory, and a wake, load all of that memory within seconds, and a server now counts each one at all of it from its start, not a quarter. A burst of copies or wakes bigger than one server's memory no longer overfills it: the ones past its room wait while it makes room, and if none comes in time answer 503 host_memory_pressure with Retry-After. Send them again after it, with the same Idempotency-Key, as the SDKs do. See errors.