fix(memoize): release flock before early return on cache hit

run_memoized() opens the lock via `exec {lock_fd}>`, a process-level fd
that is not closed when the function returns. The cache-hit path did
`source "${disk_cache_file}"; return 0` without the `flock -u` that the
memoize/write path performs before its own return, leaking the advisory
lock (and the fd) on every cache hit.

Pre-existing, but the new OFFLINE_WORK stale-serve path now takes this
early return for stale entries too (previously stale => deleted =>
fall-through to the unlocking path), so it fires far more often. A second
run_memoized() for the same key in one process would then block until
MEMOIZE_FLOCK_MAX_WAIT. Release the lock before the early return.

Assisted-by: Claude:claude-opus-4.8

Author: iav

lib/functions/general/memoize-cached.sh

@@ -97,6 +97,7 @@ function run_memoized() {
 			display_alert "Using cached" "${var_n}" "info"
 			# shellcheck disable=SC1090 # yep, I'm sourcing the cache here. produced below.
 			source "${disk_cache_file}"
+			flock -u "${lock_fd}"
 			return 0
 		fi
 	fi