Recover commits whose finalize stage times out

@snawaz is already working on this.

Since Sep 9, 2026.

## Outcome

Recover commits whose finalize stage times out.

## Scope

**Problem**
The committor can leave accounts stuck when a two-stage commit reaches base layer but the finalize stage fails or times out before the validator observes a processed status.

Observed failure shape:
- Commit stage succeeds and initializes DLP commit-state records.
- Finalize stage returns `FailedToFinalizeError` / `CannotGetTransactionSignatureStatus`.
- Rows are persisted as `FailedFinalize`.
- The runtime execution worker completes the intent and unblocks the account set instead of retrying the finalize stage.
- On restart, recovery only reloads pending rows, so `FailedFinalize` rows can remain stranded.

For affected undelegations, later direct undelegate attempts can fail because the commit-state PDA already exists, while the account was never finalized/undelegated.

**Example**
For account `CqP3QW6MtCpEuVuySshx5XzdN1JnsE5sbWQfPQhxNLi9`, the commit stage landed, but the finalize stage timed out and the committor DB recorded `FailedFinalize`. The account remained stuck.

**Expected behavior**
- `FailedFinalize` rows should be recoverable after restart.
- Runtime execution should retry the finalize-stage recovery path instead of treating the intent as done.
- Retrying a `FailedFinalize` row must not rebuild or resend the commit-stage transaction; it should reuse the recorded commit-stage state and continue with finalize/undelegate work.
- Repeated finalize retry failures should preserve `FailedFinalize` status rather than downgrading the row to a generic failed-process state.

## Acceptance criteria

- [ ] The outcome is implemented and validated.