Add SignalReason to FinishJobRequest

[walkerdb]

Problem

Stacks (e.g. agent-stack-k8s) fail jobs via POST /stacks/{key}/jobs/{uuid}/finish, but FinishJobRequest only carries exit_status and detail. That means stack-terminated jobs land in Buildkite with no machine-readable signal_reason, so:

• retry: automatic: [{signal_reason: stack_error}] — the documented way to auto-retry infrastructure failures — can't match jobs failed through this API.
• API/webhook/EventBridge/OpenTelemetry consumers can't distinguish "the stack failed to provision this job" (pod pending timeout, image pull failure, podSpec rejection) from an ordinary lost agent: both surface as failed / exit_status: -1.

Temporary-agent-based stacks used to set signal_reason: stack_error when acquiring and failing a job; the Stacks API finish path dropped that capability. agent-stack-k8s still computes the reason at every failure site (FailureInfo{Reason: agent.SignalReasonStackError}) but has nowhere to send it.

Change

Adds an optional SignalReason field (json:"signal_reason,omitempty") to FinishJobRequest — backward-compatible, omitted when unset.

If the server-side /finish endpoint doesn't yet accept this parameter, please treat this PR as the feature request for it.

Companion PR: buildkite/agent-stack-k8s#916 plumbs the already-computed reason through the controller.

Testing

go test ./... passes; the round-trip is asserted in TestFinishJob.

🤖 Generated with Claude Code