T-17170 Flush after each RQ job and start forked children with a clean handler

[PetrHeinz]

RQ runs every job in a forked work horse that ends with os._exit(), which skips the interpreter shutdown that normally lets the handler send what it has buffered. A job that finishes within the flush interval therefore loses all of its records. That is the pattern in #36: the long-lived worker's own rq.worker lines arrive, the job's own lines do not.

Two changes, tests first (the first commit is expected to fail CI):

• logtail.rq.Worker flushes every LogtailHandler after each job, inside the work horse, before RQ exits it. Use it with rq worker --worker-class logtail.rq.Worker, or RQ = {"WORKER_CLASS": "logtail.rq.Worker"} with django-rq. rq stays out of the package's dependencies, the module only imports it when used.
• A forked child now starts with a fresh queue, no flush thread and a new HTTP session. Until now it inherited the parent's queue, so once the child flushes it would re-send whatever the parent still had buffered at fork time, a thread object whose thread does not exist in the child, and a requests.Session whose socket it shared with the parent.

🤖 Generated with Claude Code