Crucible panicked after hitting IO completion error

I have some running instances returned "Failed Dependency" error when I tried to stop or reboot them (see https://github.com/oxidecomputer/omicron/issues/5705 for the context).
Looking at the propolis log for one of them, I saw that it panicked at one point:

BRM42220009 # grep IO /pool/ext/d0584f4a-20ba-436d-a75b-7709e80deb79/crypt/debug/oxz_propolis-server_5317472d-c334-451e-ad6a-7c184289bbf7/system-illumos-propolis-server:default.log.1714980545 | looker
07:23:48.646Z INFO propolis-server (crucible-fda6b60b-1ea5-4d90-8610-64e691b45879): Scrubs from block 0 to 4194304 in (256) 131072 size IOs pm:25
07:29:03.883Z ERRO propolis-server (crucible-fda6b60b-1ea5-4d90-8610-64e691b45879): IO completion error: missing 14411 
     = downstairs
    client = 2
    session_id = 023341a5-1224-42d4-a5eb-30244dae3a60

The complete log file is attached here:
system-illumos-propolis-server:default.log.1714980545.log

Start with the attached propolis log and the related oxidecomputer/omicron#5705 issue, focusing on the IO completion error reporting a missing request before the panic. Trace that failure through Crucible's IO completion handling, then define done as reproducing the panic or identifying its cause and adding a fix with regression coverage.