What's new

• Add a benchmark run API to powerpipe server: POST /api/v0/benchmark/<benchmark>/run starts a benchmark run, GET on the run returns its status, GET .../snapshot returns the same snapshot as --export=<file>.pps, and DELETE cancels and removes it. Many runs share the one loaded mod and the server's database connections, so a server running several benchmarks at once uses much less memory and time than a process for each run. At most 10 runs run at once; a run that no client has read for a minute is removed. Note: the run API has no authentication. Anyone who can reach the server can start and read benchmark runs, so use it only on a network you trust.
• Add powerpipe benchmark run <benchmark> --host <url> to run a benchmark on a powerpipe server through the run API. It writes the same snapshot and exits with the same codes as a local run. It supports --output=snapshot|none, --export=<file>.pps|pps, --benchmark-timeout and --progress, for one benchmark.

Enhancements

• powerpipe benchmark run and powerpipe control run now load only the resources the benchmark or control uses, which cuts load time and memory.
• powerpipe benchmark run --export now streams control results to the export file, and keeps them on disk when the output does not print them, so memory no longer grows with the number of result rows.

Dependencies

• Bump pipe-fittings/v2 to v2.9.5, github.com/mattn/go-sqlite3 to v1.14.52, github.com/spf13/cobra to v1.10.2 and github.com/spf13/viper to v1.21.0.
• Bump go.opentelemetry.io/otel to v1.45.0 and github.com/klauspost/compress to v1.20.3.