Fix remaining-task semantics for completed parent chains

[deverman]

Summary

Fix remaining/default task semantics so children of completed or dropped parent tasks are no longer treated as remaining.

Change

• update bridge remaining-state helpers to respect the full parent task chain
• update the embedded JXA path to apply the same ancestor-aware remaining logic
• add live integration coverage for children under completed parents in bridge and JXA paths

Why

Issue #27 is about read trust before write support. A task should not appear actionable if an ancestor task has already been completed or dropped.

Validation

• swift test
• benchmark-gate-check --tool task-counts
• benchmark-gate-check --tool list-tasks
• 10-minute smoke benchmark for list-tasks
• 10-minute smoke benchmark for task-counts

Notes

The semantic gates are slow on this machine and database, but they passed cleanly once run with a sufficient timeout budget.

[deverman]

Merge-order note: this PR is intentionally independent from issue #28. Issue #28 was branched from origin/master and does not require #37 to merge first. Merge this PR whenever the read-trust fix is reviewed and ready; do not block #28 on this unless the write-contract review explicitly decides to reference the new completed-parent semantics examples.

[deverman]

Validation after merging current origin/master into issue-27-read-trust-parity:

• node --check Plugin/FocusRelayBridge.omnijs/Resources/BridgeLibrary.js
• swift test (68 tests passed; existing MCP SDK deprecation warnings only)
• ./scripts/install-plugin.sh; fully restarted OmniFocus
• ./.build/debug/focusrelay bridge-health-check
• swift run focusrelay benchmark-gate-check --tool task-counts
• swift run focusrelay benchmark-gate-check --tool list-tasks
• swift run focusrelay benchmark-gate-check --tool project-counts
• swift run focusrelay benchmark-task-counts --duration-hours 0.1667 --warmup-calls 10 --interval-ms 2000 --cooldown-ms 3000 --output-dir /tmp/focusrelay-issue27-task-counts-smoke
• swift run focusrelay benchmark-list-tasks --duration-hours 0.1667 --warmup-calls 10 --interval-ms 2000 --cooldown-ms 3000 --output-dir /tmp/focusrelay-issue27-list-tasks-smoke
• swift run focusrelay benchmark-project-counts --duration-hours 0.1667 --warmup-calls 10 --interval-ms 2000 --cooldown-ms 3000 --output-dir /tmp/focusrelay-issue27-project-counts-smoke

Smoke results for all three affected tools: 0 mismatches, 0 errors, 0 timeouts, 0 timeout diagnostic entries.

Merge order note: this remains the bottom PR in the mutation/read-trust stack and should merge before #38, #40, #41, #42, #43, #44, #46, #49, #50, #51, #52, and #54.