Test flake: silent failure to initialize cockroachdb?

https://github.com/oxidecomputer/omicron/pull/3887/checks?check_run_id=15946171931 failed; the immediate symptom in the deploy job a timeout after trying to log in for 10 minutes:

673	2023-08-16T15:51:53.331Z	2023-08-16 15:51:52.586682700 UTC: login failed: logging in: error sending request for url (https://recovery.sys.oxide.test/v1/login/recovery/local): error trying to connect: operation timed out
674	2023-08-16T15:51:54.332Z	2023-08-16 15:51:53.587485600 UTC: attempting to log into API
675	2023-08-16T15:52:09.366Z	2023-08-16 15:52:08.621092329 UTC: login failed: logging in: error sending request for url (https://recovery.sys.oxide.test/v1/login/recovery/local): error trying to connect: operation timed out
676	2023-08-16T15:52:10.368Z	2023-08-16 15:52:09.622786274 UTC: attempting to log into API
677	2023-08-16T15:52:25.402Z	2023-08-16 15:52:24.656701751 UTC: login failed: logging in: error sending request for url (https://recovery.sys.oxide.test/v1/login/recovery/local): error trying to connect: operation timed out
678	2023-08-16T15:52:26.403Z	2023-08-16 15:52:25.657578343 UTC: attempting to log into API
679	2023-08-16T15:52:41.436Z	2023-08-16 15:52:40.691037367 UTC: login failed: logging in: error sending request for url (https://recovery.sys.oxide.test/v1/login/recovery/local): error trying to connect: operation timed out
680	2023-08-16T15:52:42.439Z	Error: logging in
681	2023-08-16T15:52:42.464Z	
682	2023-08-16T15:52:42.489Z	Caused by:
683	2023-08-16T15:52:43.040Z	    timed out after 609.325828065s

All three nexus logs have many warning that seem to indicate the database wasn't set up correctly, I think?

WARN	nexus: Cannot look up rack: Object (of type ById(24ed7902-3649-4e0e-8635-cd082ae9b0c0)) not found: rack
    file = nexus/src/app/rack.rs:555

but the sled-agent logs do not report any issues with initial CRDB setup:

328	2023-08-16T15:43:19.640Z	INFO	SledAgent (ServiceManager): Formatting CRDB
    file = sled-agent/src/services.rs:2248
...
330	2023-08-16T15:43:22.432Z	INFO	SledAgent (ServiceManager): Formatting CRDB - Completed
    file = sled-agent/src/services.rs:2269

Start with the failed deploy check from PR 3887 and compare the nexus warnings at nexus/src/app/rack.rs:555 with CRDB formatting in sled-agent/src/services.rs:2248 and :2269. Reproduce the deploy-job login timeout and trace whether initialization is incomplete; done means identifying the failure cause and adding a reliable regression check or fix.