fix: continue Pi session for verdict repair

[ruby-dlee]

Intent

Make Azure Crosscheck recover a Pi output-length or malformed verdict attempt by continuing the same isolated exact-head review session for one short verdict-only retry, while preserving cleanup, telemetry, and fail-closed semantics.

What Changed

• Continue malformed or length-truncated Pi verdict retries in the same isolated review session with a minimal-effort, verdict-only prompt.
• Preserve combined attempt telemetry and fail closed after one unsuccessful repair.
• Extend Azure Crosscheck coverage for session reuse and length recovery, and make the Herdr prune-safety test use bounded heartbeat polling.

Risk Assessment

✅ Low: The change is narrowly scoped and preserves isolation, bounded single-retry behavior, telemetry aggregation, cleanup, and fail-closed handling while continuing the same Pi session for verdict repair.

Testing

The successful baseline was supplemented with focused executable Pi CLI simulations demonstrating same-session recovery from malformed and output-length attempts, cumulative telemetry, preserved isolation, and fail-closed behavior after exactly one repair; all checks passed and the worktree remained clean.

Evidence: Pi verdict repair and fail-closed transcript

PI RUNTIME repairs one verdict-protocol miss, then fails closed PASS the digest-bound Pi runtime bounds verdict repair and remains fail closed

PI RUNTIME repairs one verdict-protocol miss, then fails closed
PASS the digest-bound Pi runtime bounds verdict repair and remains fail closed

- Outcome: 🔧 1 issue found → auto-fixed ✅ across 2 runs (30m7s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

🔧 **Test** - 1 issue found → auto-fixed ✅

• 🚨 tests failed with exit code 1
• if [ "${FM_AZURE_VALIDATION_CELL:-0}" = 1 ]; then exec "$FM_AZURE_VALIDATION_SHARD_BRIDGE" behavior --count "${FM_AZURE_VALIDATION_SHARD_COUNT:-8}"; else exec bin/fm-no-mistakes-test-command.sh; fi

🔧 Fix: Stabilize Herdr heartbeat readiness test
✅ Re-checked - no issues remain.

• if [ "${FM_AZURE_VALIDATION_CELL:-0}" = 1 ]; then exec "$FM_AZURE_VALIDATION_SHARD_BRIDGE" behavior --count "${FM_AZURE_VALIDATION_SHARD_COUNT:-8}"; else exec bin/fm-no-mistakes-test-command.sh; fi
• Configured baseline command (reported as already successful): if [ "${FM_AZURE_VALIDATION_CELL:-0}" = 1 ]; then exec "$FM_AZURE_VALIDATION_SHARD_BRIDGE" behavior --count "${FM_AZURE_VALIDATION_SHARD_COUNT:-8}"; else exec bin/fm-no-mistakes-test-command.sh; fi
• Executed the focused pi_reviewer_runtime_unit harness against bin/fm-crosscheck-pi-reviewer.py, covering valid submission, internal retry, malformed verdict recovery, same-session continuation, minimal-effort verdict-only retry, telemetry aggregation, and bounded failures.
• Repeated the focused executable harness with length included among exhausted-repair scenarios, proving both length-stop recovery and failure after one unsuccessful retry.
• Verified git status --short remained clean after testing.

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.