[codex] Harden heartbeat scheduling and runtime controls (#4223)

## Thinking Path

> - Paperclip orchestrates AI agents through issue checkout, heartbeat
runs, routines, and auditable control-plane state
> - The runtime path has to recover from lost local processes, transient
adapter failures, blocked dependencies, and routine coalescing without
stranding work
> - The existing branch carried several reliability fixes across
heartbeat scheduling, issue runtime controls, routine dispatch, and
operator-facing run state
> - These changes belong together because they share backend contracts,
migrations, and runtime status semantics
> - This pull request groups the control-plane/runtime slice so it can
merge independently from board UI polish and adapter sandbox work
> - The benefit is safer heartbeat recovery, clearer runtime controls,
and more predictable recurring execution behavior

## What Changed

- Adds bounded heartbeat retry scheduling, scheduled retry state, and
Codex transient failure recovery handling.
- Tightens heartbeat process recovery, blocker wake behavior, issue
comment wake handling, routine dispatch coalescing, and
activity/dashboard bounds.
- Adds runtime-control MCP tools and Paperclip skill docs for issue
workspace runtime management.
- Adds migrations `0061_lively_thor_girl.sql` and
`0062_routine_run_dispatch_fingerprint.sql`.
- Surfaces retry state in run ledger/agent UI and keeps related shared
types synchronized.

## Verification

- `pnpm exec vitest run
server/src/__tests__/heartbeat-retry-scheduling.test.ts
server/src/__tests__/heartbeat-process-recovery.test.ts
server/src/__tests__/routines-service.test.ts`
- `pnpm exec vitest run src/tools.test.ts` from `packages/mcp-server`

## Risks

- Medium risk: this touches heartbeat recovery and routine dispatch,
which are central execution paths.
- Migration order matters if split branches land out of order: merge
this PR before branches that assume the new runtime/routine fields.
- Runtime retry behavior should be watched in CI and in local operator
smoke tests because it changes how transient failures are resumed.

> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.

## Model Used

- OpenAI Codex, GPT-5-based coding agent runtime, shell/git tool use
enabled. Exact hosted model build and context window are not exposed in
this Paperclip heartbeat environment.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [ ] If this change affects the UI, I have included before/after
screenshots
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] I will address all Greptile and reviewer comments before
requesting merge

packages/adapters/codex-local/src/server/parse.ts

@@ -1,5 +1,9 @@
 import { asString, asNumber, parseObject, parseJson } from "@paperclipai/adapter-utils/server-utils";
 
+const CODEX_TRANSIENT_UPSTREAM_RE =
+  /(?:we(?:'|’)re\s+currently\s+experiencing\s+high\s+demand|temporary\s+errors|rate[-\s]?limit(?:ed)?|too\s+many\s+requests|\b429\b|server\s+overloaded|service\s+unavailable|try\s+again\s+later)/i;
+const CODEX_REMOTE_COMPACTION_RE = /remote\s+compact\s+task/i;
+
 export function parseCodexJsonl(stdout: string) {
   let sessionId: string | null = null;
   let finalMessage: string | null = null;
@@ -71,3 +75,25 @@ export function isCodexUnknownSessionError(stdout: string, stderr: string): bool
     haystack,
   );
 }
+
+export function isCodexTransientUpstreamError(input: {
+  stdout?: string | null;
+  stderr?: string | null;
+  errorMessage?: string | null;
+}): boolean {
+  const haystack = [
+    input.errorMessage ?? "",
+    input.stdout ?? "",
+    input.stderr ?? "",
+  ]
+    .join("\n")
+    .split(/\r?\n/)
+    .map((line) => line.trim())
+    .filter(Boolean)
+    .join("\n");
+
+  if (!CODEX_TRANSIENT_UPSTREAM_RE.test(haystack)) return false;
+  // Keep automatic retries scoped to the observed remote-compaction/high-demand
+  // failure shape; broader 429s may be caused by user or account limits.
+  return CODEX_REMOTE_COMPACTION_RE.test(haystack) || /high\s+demand|temporary\s+errors/i.test(haystack);
+}