#282 periodic heartbeat to prevent Docker idle timeout

[VasilevNStas]

Long-running judges update sessions can exceed Docker's stdout-idle timeout, causing the container to be killed silently.

What this PR does

Emits a timestamped [heartbeat] line to stdout every HEARTBEAT_INTERVAL seconds (default 60) while judges runs:

[heartbeat] Job 79196 is still running at 2026-06-29T20:15:30Z
[heartbeat] Job 79196 is still running at 2026-06-29T20:16:30Z

A trap on EXIT ensures the background heartbeat process is always cleaned up — even when judges fails via set -e.

Configuration

Env var	Default	Description
HEARTBEAT_INTERVAL	60	Seconds between heartbeat lines

Checklist

• bundle exec rubocop — 0 offences
• bundle exec rake — all tasks pass
• HoC ≤ 133

@yegor256 please review

[VasilevNStas]

Why this matters

Docker kills containers that produce no stdout for a configurable idle time (commonly 5–10 minutes). A judges update that's genuinely working — e.g. processing a large factbase or waiting on external API calls — can easily exceed this threshold.

The symptom is silent: the container disappears, baza.rb sees a timeout, and nobody knows whether the job finished, crashed, or was murdered by Docker.

After this PR, a heartbeat line arrives every 60s:

[heartbeat] Job 79196 is still running at 2026-06-29T20:16:30Z

Three benefits:

1. Docker keeps the container alive — stdout activity resets the idle timer
2. Operators see the job making progress — the timestamp advances, proving the process isn't wedged
3. No additional dependencies — pure bash while/sleep, no need for ts or pv

For jobs where 60s is too aggressive (or not aggressive enough), HEARTBEAT_INTERVAL is configurable per environment.