feat: collect performance metrics for regenerating cached suggestions

[DarshanCode2005]

Related to #1002

Problem

Performance-sensitive batch jobs (cache regeneration, CVE ingestion, garbage collection etcetra) only log durations and counts to stdout/logs. Operators have no Grafana view of whether these jobs are slowing down or processing less data over time. The existing SQL exporter gives aggregate counts and recency (sql_delta, sql_matching), but not job durations or throughput.

What this PR does

Establishes the end-to-end wiring for application-level metrics from batch jobs to Grafana. Per issue scope, this is wiring + one sample metric, not full instrumentation.

Pipeline

regenerate_cached_suggestions -> cache_regeneration.prom -> node_exporter textfile collector -> Prometheus -> Grafana

[Erethon]

I've left an early review. Besides the inline comments, can you also fix the python formatting so that CI is happy please?

[fricklerhandwerk]

This needs to go to the other settings = -- sorry, this settings jungle is a mess, I'm in the process of cleaning it up so this sort of friction doesn't occur in the future #1166

[fricklerhandwerk]

LGTM now! @Erethon final words?

[fricklerhandwerk]

Apart from formatting this seems correct now. I can't merge it yet because I ran into an unrelated problem introduced by #1130, which makes it impossible to regenerate the cache and try it out the metrics locally.

[DarshanCode2005]

Took a while to debug , but did it : )

The issue was that after @fricklerhandwerk refactored write_metrics_textfile() to accept a filename (name: str) and use METRICS_TEXTFILE_DIR from settings, the tests weren't updated. They were still passing a full Path object (tmp_path / "sample"). Since tmp_path is an absolute path, pathlib ignored METRICS_TEXTFILE_DIR entirely, so the tests wrote directly to the temporary directory instead of using the configured metrics directory. As a result, the tests still passed but weren't actually verifying the production behavior.

To fix this, the tests now use override_settings(METRICS_TEXTFILE_DIR=tmp_path) so they're isolated and don't depend on the environment. They also pass simple string filenames like "sample" and "cache_regeneration", which matches how the function is used in production, for example in regenerate_cached_suggestions.py. The implementation of src/shared/metrics.py itself didn't change. Only the tests were updated to correctly reflect the refactor.

done

[fricklerhandwerk]

Unfortunately I can't test this locally for dumb reasons (related #1045 (comment)), but this is well-tested and I trust it will work and at worst need small fixups in the future.