fix: warn instead of silently suppress pid_ts write failure in acquire_boilerplates_lock

[kitsuyui]

Summary

acquire_boilerplates_lock() writes pid_ts after acquiring the lock so that waiting processes can detect a stale lock (owner died via SIGKILL). The write was previously suppressed with 2>/dev/null || true, meaning a failure (disk full, quota, permissions) would silently disable stale detection without any visible signal.

With this change, if the write fails the action logs a warning and continues. The lock is already held (the mkdir succeeded), so operation proceeds, but the degraded state is visible in the action log for diagnosis.

Change

- printf '%s %s\n' "$$" "$(date +%s)" > "${lock_dir}/pid_ts" 2>/dev/null || true
+ if ! printf '%s %s\n' "$$" "$(date +%s)" > "${lock_dir}/pid_ts"; then
+   echo "install-gibo: warning: failed to write lock metadata (${lock_dir}/pid_ts); stale lock detection will be unavailable" >&2
+ fi

Impact

• No functional change under normal conditions.
• On write failure: previously silent; now emits a warning to stderr and continues.
• The polling loop's stale-detection path ([ -f "${lock_dir}/pid_ts" ]) still behaves the same — it simply won't fire if the file was never created.
• A SIGKILL-killed lock holder with a missing pid_ts still causes a 300-second wait; this PR makes that scenario diagnosable rather than mysterious.

Test plan

• CI passes on all platforms (ubuntu, macos, windows)
• No change in behavior when pid_ts write succeeds (normal path)