feat(nodetask): consume gov task results — outputs, conditions, pending loop

[bdchatham]

The controller half of the SeiNodeTask governance UX work (design: sei-protocol/bdchatham-designs#98) — consumes the seictl v0.0.61 gov completion contract. This is the payoff: a completed gov task now surfaces its proposal ID / tx hash, a task reports Failed instead of a false Complete when a tx doesn't commit, and the integration test's brittle proposal-ID REST scan is deleted.

Depends on: seictl#219 (v0.0.61). Follows PR #450 (stop-resubmit).

Changes

• go.mod → seictl v0.0.61.
• sidecarExecution.Result() captures the terminal task result (both success and failure paths).
• driveTask decodes the gov result and branches: committed_ok → Complete + Ready/Confirmed + outputs; committed_failed → Failed/TxFailed + outputs; pending → reset to Pending and re-submit (same task ID → engine run+1 → marker adopt re-checks), bounded by spec.timeoutSeconds; on timeout → Failed/InclusionUndetermined (txHash retained). Non-gov failures unchanged.
• populateOutputs stamps proposalId/txHash/height into the existing CRD Outputs (decode-by-kind, via a local wire-mirror — no heavy sidecar-tasks import). No CRD schema change.
• SDK TaskOutputs gains gov arms; integration test reads the proposal ID off the task output and deletes resolveProposalID (the plan-name REST scan).
• Refreshed the stale "outputs deferred / chain-as-medium" docs.

Tests

controller_gov_test.go: Confirmed / CommittedFailed / Pending→resubmit→converge / PendingTimeout / GovVote-confirmed / no-result→generic-failure, plus TestGovResultMirrorMatchesWire (guards the local mirror against seictl GovTxResult drift).

Sign-off

systems-engineer (pending loop, decode, idempotency — logic correct, no changes) and kubernetes-specialist (condition discipline, single-patch, contract-safe) — both green.

⚠️ Deploy-order (one-way-door adjacent)

Roll this controller before the seictl v0.0.61 sidecar image. A pre-2b controller treats the new sidecar's pending→Failed as terminal and would prematurely fail an in-flight gov tx. New-controller + old-sidecar is graceful (nil result → today's behavior).

Reviewer flag for confirmation

Gov-kind success now emits Ready reason=Confirmed (was TaskComplete). Confirm no PromQL/runbook filters the old reason for gov kinds (gov-ops is a newer surface, likely safe).

🤖 Generated with Claude Code

[cursor]

PR Summary

High Risk
Gov reconciliation and on-chain tx outcome handling changed materially; deploy order requires this controller before seictl v0.0.61 sidecars or in-flight gov tasks can fail prematurely, and gov success now uses Ready reason Confirmed instead of TaskComplete.

Overview
Bumps seictl to v0.0.61 and wires the controller to the sidecar’s structured gov task results (txHash, proposalId, inclusionStatus). Terminal GetTask payloads are captured on sidecarExecution.Result() and decoded via a local govResult mirror in govoutputs.go.

Gov completion semantics replace a flat Complete/Failed poll: committed_ok → Complete with CRD outputs and Ready reason Confirmed; committed_failed → Failed TxFailed with outputs; pending resets task status to Pending and re-submits (same task ID) until spec.timeoutSeconds, then Failed InclusionUndetermined. Execution timeout logic is factored into execTimedOut.

The SDK exposes gov fields on TaskOutputs and translateTaskOutputs; package docs drop the “chain-as-medium only” stance for completed gov tasks. The integration upgrade test reads proposalID from task output and removes the resolveProposalID REST scan.

New controller_gov_test.go covers confirmed, tx failed, pending re-submit/converge, pending timeout, wire-shape drift guard, and generic failures without a result payload.

^{Reviewed by Cursor Bugbot for commit faa59a0. Bugbot is set up for automated code reviews on this repo. Configure here.}