fix(demo): wait for full ledger coverage before log dump in FVV demo

[sei-ahouseholder]

• Fixes bug: FVV demo occasionally fails test_invariant_8_late_joiner_has_full_history (intermittent backfill miss) #1363

Summary

_phase_case_closure in fvv_demo.py waited only for the tail entry
hash before dumping JSONL logs. Announce(CaseLedgerEntry) activities
are delivered independently per-entry; an intermediate entry (e.g.
logIndex=17) can arrive after the tail, so _phase_dump_case_ledgers
captured a gapped log, causing test_invariant_8_late_joiner_has_full_history
to fail intermittently.

Adds wait_for_contiguous_ledger_coverage() to polling.py which polls
until the replica holds all log indices 0…expected_tail_index before
the dump proceeds. Replaces both wait_for_finder_log_entry() call sites
in the FVV demo (Phase 2 sync verification and Phase 6 case closure) with
this stronger gate.

Changes

• vultron/demo/helpers/polling.py: add wait_for_contiguous_ledger_coverage()
• vultron/demo/helpers/__init__.py: re-export new helper
• vultron/demo/scenario/fvv_demo.py: replace tail-hash-only wait with
  contiguous coverage check in _phase_sync_verification and _phase_case_closure
• test/demo/test_fvv_demo.py: 5 regression tests in TestWaitForContiguousLedgerCoverage

Verification

• 5 new regression tests — including the exact race (tail present, intermediate entry absent)
• 4706 unit tests pass (uv run pytest --tb=short)
• Black, flake8, mypy, pyright all clean

[sei-ahouseholder]

PR Review: #1433 — FVV late-joiner backfill fix

Phase 1 — Issue Linkage: PASS

• Closes #1363 ✓
• Implementation scope matches the issue exactly: replace tail-hash-only wait
  with full contiguous index coverage.

Phase 2 — PR Body Format: PASS

• Closing reference at top ✓
• Summary, Changes, Verification sections present ✓
• Test count (5 new) is a real number ✓
• Co-authored-by trailer present in the fix commit ✓

Phase 3 — Domain Context / Spec Conformance: FIXED

The new wait_for_contiguous_ledger_coverage() docstring originally cited
SYNC-02-003 — but that spec governs replication originating from the
CaseActor, not the contiguous-coverage pre-dump gate. The correct reference
is SYNC-10-004 ("catch-up gate MUST require a contiguous canonical log
prefix from genesis").

FIXED (commit c9644a27): corrected docstring spec reference to
SYNC-10-004.

Phase 4 — ADR Check: PASS

No new architectural layers, protocols, or public API changes; no ADR required.

Phase 5 — AGENTS.md Compliance: PASS

Demo helper changes only; no BT nodes, use cases, or core domain touched.

Phase 6 — Code Review: PASS

• wait_for_contiguous_ledger_coverage builds expected_indices once before
  the _check closure (efficient set comparison). ✓
• Uses _poll_until generic primitive consistently with all other helpers. ✓
• Filters by case_id to avoid counting entries from other cases. ✓
• Logs missing indices (up to 10) at DEBUG for diagnosability. ✓
• fvv_demo.py: both Phase 2 and Phase 6 sites updated; the Phase 2 site
  also correctly checks Vendor2 in addition to Finder. ✓

Phase 7 — Notes Currency: PASS

notes/sync-ledger-replication.md covers the design but doesn't need
updating — the change is a demo-layer polling improvement, not an
architecture change.

Phase 8 — Test Coverage: PASS

5 regression tests in TestWaitForContiguousLedgerCoverage cover:

1. Happy path (all indices present) ✓
2. Missing intermediate index — exact #1363 race ✓
3. Tail present but early entries absent ✓
4. Entries from other cases excluded ✓
5. No entries present ✓

Phase 9 — CI Status: PASS

All 12 checks pass, including FVV Demo Integration and Two-Actor Demo Integration. ✓

DEFER

#1434 — two_actor_demo.py still uses wait_for_finder_log_entry()
(tail-hash-only) at lines 574 and 770. The same race applies to the
two-actor demo. Tracked in follow-up issue #1434; out of scope for this
PR since the bug report was FVV-specific and two-actor CI is currently green.

---

Overall: ready to merge. Root cause fixed, 5 regression tests, one
inaccurate spec reference corrected in commit c9644a27.