[Bug] Driver silently ignores an out-of-band apply failure (agent encodes it as phase "apply" = 0x00)

[traviswu-bigstack]

Description

When snapshot_apply fails on a node, the OS-phase agent reports it out-of-band with writeSEL("apply", "error", err) (cmd/phone-home-agent/main.go:634). But "apply" is not in the agent's phaseCode map, so it silently encodes to byte 0x00. The driver's selPhaseName has no 0x00 entry → Phase="" → selState returns "" → the record is ignored. The deploy UI stays stuck on "applying" instead of showing the failure. In-band reportApplyFailed only covers it while mgmt is reachable — which post-apply air-gap usually isn't, and OOB is exactly the fallback for that case.

Observed live on qa41-r630 (2026-09-01): apply failed, the node's BMC SEL held 00 ff …"snap" (phase 0x00, result 0xff=error), and the driver showed apply active indefinitely.

How to Reproduce

1. Deploy a node where hex_config snapshot_apply returns a hard error.
2. The node writes the apply-error SEL record; the in-band report doesn't reach the driver.
3. Driver UI sits on "applying"; the node never shows errored.

Expected behavior

An out-of-band apply failure surfaces as an errored node.

Additional Context

Root cause is a bad phase string in the agent; the driver's OOB decode is also not defensive about the error result. Fix both: the agent emits a real error phase, and the driver treats any cube record whose result is error as terminal regardless of the phase byte (only the apply-failure record carries the error result, so it can't misfire).

Output artifacts (Definition of Done)

• Handbook knowledge update — fold into the existing OOB-SEL known-issue or add a short note.