Harden daemon summarization single-writer semantics against crash-restarts

Context

Surfaced during the PR #10967 (migrate summary sweeps to daemon) review by @neo-opus-4-7. While PR #10967 successfully moved the summarization lifecycle to bridge-daemon.mjs, it uses an in-process boolean flag (summarizationRunning) to prevent concurrent summarize-sessions.mjs invocations.

The Problem

The in-process summarizationRunning mutex is not crash-safe. If the bridge-daemon process itself dies mid-spawn (e.g., SIGKILL while the summarize-sessions.mjs child is running), it leaves an orphaned child process. On daemon restart, summarizationRunning re-initializes to false, and the new daemon could spawn a second summarize-sessions.mjs child. This defeats the single-writer guarantee.

The Fix

Introduce a true crash-safe locking mechanism for the summarization child process (e.g., a summarization.lock file or explicit PID-tracking mechanism in checkSummarizationLifecycle()).

Acceptance Criteria

• Investigate and implement a crash-safe lock for summarize-sessions.mjs spawns.
• Ensure that an orphaned child process does not allow a restarted daemon to spawn a concurrent summarization job.
• Add or update tests to verify crash-recovery logic.