Fix flaky unit test on merge-triggered CI runs (#7)

- entrypoint.sh: bound the target connectivity check with `-w 2` (the only unbounded step in the startup path)
- tests/entrypoint.bats: widen wait_for_log's budget from 5s to 10s

Merge-triggered runs land on a colder runner (Docker cache evicted by other repos' scheduled jobs since the PR was last tested) than PR-check runs, and the first backgrounded entrypoint.sh in the suite was occasionally landing just past the old 5s window — always passed on rerun once warm.

tests/entrypoint.bats

@@ -58,7 +58,11 @@ socat_pid() {
 wait_for_log() {
     pattern="$1"
     tries=0
-    while [ "$tries" -lt 20 ]; do
+    # 40 * 0.25s = 10s — merge-triggered runs can land on a cold runner
+    # (evicted Docker cache, first container start), pushing the very first
+    # backgrounded entrypoint.sh past a tighter budget even though nothing's
+    # actually wrong; a rerun on a warm runner always passes.
+    while [ "$tries" -lt 40 ]; do
         grep -qE "$pattern" "$LOG" 2>/dev/null && return 0
         tries=$((tries + 1))
         sleep 0.25