Cache inspection API

peryx serve keeps the metadata store open for its process lifetime. redb refuses a second open while that writer exists, so the offline peryx cache commands need the server to stop first. The cache inspection API reads through the server's existing handle.

Each endpoint requires administration:read and a local administrator credential. They return text/plain; charset=utf-8 with the same tab-separated rows as the matching offline command.

GET /+cache matches peryx cache list. GET /+cache/size matches peryx cache size, and GET /+cache/fsck matches peryx cache fsck. There is no endpoint for peryx cache repair: these endpoints read, and a rebuild writes.

GET /+cache accepts the offline list filters as query parameters: index, resource, digest, stale, min_age_secs, and min_size_bytes.

curl --user administrator:password 'https://packages.example/+cache?index=pypi&stale=true'
curl --user administrator:password https://packages.example/+cache/size
curl --user administrator:password https://packages.example/+cache/fsck

The handlers run filesystem scans outside the async request executor. Clients can write cache entries during an inspection, so one report can span two committed states.

Operators can use the offline commands when a server cannot start. Those commands do not build a serving state and require exclusive access to the metadata store.