Make the daily scraper survive refused connections from govjobs.publi…

[masterries]

…c.lu

Recent runs died with an uncaught "[Errno 111] Connection refused" in the middle of the listing crawl, so nothing was saved, committed or notified. The site refuses TCP connections when a runner IP opens too many of them, and every bare requests.get() opened a new TCP+TLS connection.

job_scraper.py:

• one keep-alive requests.Session for all requests (the live test ran 41 requests over a single connection), timeouts, urllib3 retries with backoff for refused connections and 429/5xx (Retry-After capped at 60s)
• crawl the listing first, then the missing detail pages; on errors stop and save what was collected (::warning::), exit 1 only if nothing was scraped; missing details are fetched on the next run
• circuit breaker after 3 consecutive failed detail pages, max 100 detail fetches and a 20 minute time budget per run
• check listing status codes, dedupe links, max page guard, compare the collected count with the count shown on the site

telegram.py:

• HTML parse mode, split messages between job blocks instead of every 4000 characters (cut links made Telegram reject the chunk)
• timeout, retries, exit 1 when a message can't be sent

Workflow:

• checkout@v7 / setup-python@v7 (Node 24), Python 3.14, pip cache, install from requirements.txt (drop unused firebase-admin)
• cron at 05:17 and 15:47 UTC instead of 00:00, concurrency group, timeout-minutes, contents: write, push with rebase retry
• notify and commit only on main, notify before committing so a failed notification is resent by the next run