ci: shard behavior tests by measured duration

[ruby-dlee]

Intent

Refresh PR 56 onto current main without rewriting its gate-tracked history, preserve its behavior-test sharding work, fix all current-head CI defects, and get the existing PR fully green without merging it.

What Changed

• Split behavior tests into eight deterministic, duration-balanced CI shards with isolated temp roots and per-shard timeouts.
• Added execution manifests and a final completeness guard that verifies every planned test ran exactly once and reports recorded failures.
• Partitioned the account-routing, report-stack, and teardown suites, and documented the current 87-test inventory, measured timings, and shard maintenance workflow.

Risk Assessment

✅ Low: The follow-up accurately refreshes the shard inventory, load table, and indivisible-test floor, while the full change remains well-bounded and preserves the sharding and merge-history requirements.

Testing

• ⏭️ Test - skipped

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

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

• ⚠️ docs/ci-behavior-shards.md:21 - The documented “checked-in plan” is stale after the teardown split and current-main refresh. The actual 87-test plan has loads of 475500/436053/436051/436048/436056/436296/436222/436053 ms, while this table totals only 84 tests; line 38 also names the removed fm-teardown.test.sh. Refresh the assignment table and floor description so the operational documentation matches the committed inventory.

🔧 Fix: Refresh current behavior shard plan documentation
✅ Re-checked - no issues remain.

⏭️ **Test** - skipped

Step was skipped.

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

[ruby-dlee]

Closing this temporary no-mistakes alias PR after transferring its gated head and generated pipeline record to the intended PR #56; PR #56 is running current-head CI.