🐛 Bug Fixes
- fix: retry transport failures instead of killing the run
- PR: #557
A single transport failure — fetch failed, i.e. DNS, ECONNRESET or a timeout — aborted a whole run. A real run died at list 29 of 31 on the first mdblist search, losing the two remaining lists and leaving the 29th amputated.
Both REST adapters now retry a transport failure instead of failing the run: mdblist gains a retry loop it never had (3 attempts, 1s/2s/4s backoff), and Floppy's existing 5xx retry now covers transport failures too (250ms/500ms/1s). List creation is deliberately never retried — it is not idempotent — and recovers by re-reading the list by name instead, so a network blip can no longer produce a duplicate list.