fix(ingester): break main loop when all workers exit unexpectedly

[tales-aparecida]

The main loop waited for the queue to drain (while not
process_queue.empty()), which blocks forever when a worker hangs and
never consumes its poison pill. Now the loop checks whether any worker
is still alive after each progress report and breaks with an error log
if all workers have exited while items remain in the queue.

After joining, non-zero exit codes are logged and a Prometheus counter
(kcidb_ingester_worker_failures) is incremented with reason="exception"
or reason="signal". The reason label is more actionable for alerting
(e.g. rate(worker_failures{reason="signal"}) > 0 catches OOM kills)
while the exact exit code in the log line provides detail for
debugging.

Also adds docs/ingester.md documenting the parallel ingestion
architecture, worker-queue protocol, and error handling.

Fixes: #2041

Depends on #2042

Assisted-by: Claude Opus 4.6 noreply@anthropic.com
Signed-off-by: Tales da Aparecida tales.aparecida@redhat.com

[mentonin]

commit message says alive checks run "after each progress report", but they run every loop instead

Also, commit message claims this fixes #2041 and deals with hanging, but this fix does neither - it deals with all workers dying (not hanging) before the queue is dealt with, and hanging workers still block forever (#2041 only fixes workers hanging on http requests). We can go ahead with the merge, but live workers hanging might come back as an issue later

[tales-aparecida]

The rest of the feedback I agreed, and fixed.