Handle :EXIT messages in ingestion WriteBuffer

[SAY-5]

Changes

Plausible.Ingestion.WriteBuffer calls Process.flag(:trap_exit, true) so terminate/2 can flush the buffer on shutdown, but handle_info/2 only matched :tick. When a linked port or process (the ClickHouse connection socket) exits, the GenServer receives {:EXIT, port, :normal} and crashes with a FunctionClauseError, then restarts. This adds a handle_info/2 clause that ignores :EXIT messages so the buffer keeps running. Fixes #6382.

[CLAassistant]

All committers have signed the CLA.

[aerosol]

Thanks for your contribution. With regards to the test, it seems a bit pointless to send a message to an artificially spawned buffer - you could do that against the primary buffer as well, since nothing is expected to happen - except no crash. You could experiment with triggering sending the actual EXIT message via Process.exit if you like the test to be more meaningful.

But the PR states that this is ClickHouse socket dying - in which case terminate will try to flush with a dead connection, which may lead to another crash, right?

[SAY-5]

Good points, thanks. I reworked the test to spawn a process, link it to the buffer, and kill it, so the buffer now receives a genuine {:EXIT, from, reason} through the trap-exit link rather than a hand-crafted send/2. Confirmed it fails with the FunctionClauseError from #6382 without the new clause and passes with it.

On the second point: you're right that the socket dying is a separate failure mode. The buffer isn't linked to the ClickHouse connection (IngestRepo.query! checks a connection out of the pool per call), so what #6382 actually reports is the {:EXIT, port, :normal} reaching an unmatched handle_info. A flush over a dead connection during terminate/2 would raise, but that path is independent of this crash and would be its own change, so I kept this PR scoped to swallowing the stray :EXIT. Happy to open a follow-up for the terminate/flush resiliency if you'd like it handled here.