nixception v0.3.0

nixception 0.3.0

This release focuses on timing and cache observability — making it clear
where time actually goes and how much the Nix cache is saving.

Highlights

• Runner-phase timing. The runner now measures its own setup, task, and
  wrap-up phases and reports them back, so the previously opaque
  "command execution" span is broken down into the time spent in Nix (the
  nix→runner latency: daemon scheduling and sandbox setup) versus the
  runner's setup, the task itself, and wrap-up.

• Honest wall-clock, parallelism, and throughput. The summary no longer
  mislabels the sum of per-action spans as wall-clock. It now reports the real
  elapsed wall-clock separately from cumulative action time, and adds average /
  peak parallelism and throughput (actions per second).

• Cache accounting. Actions served from the Nix cache are now distinguished
  from executed ones. The summary reports the cache hit ratio, the estimated
  time the cache saved, and the resulting speedup.

• Three-section summary. The timing report is reorganized into
  Preparation (common to all actions), Execution (executed actions
  only, so cache hits don't dilute the real-work averages), and Cached
  (hits, with the benefit estimate).

See the changelog for the full list, including the earlier
0.2.x fixes carried into this release.

Licensing

The nixception sources are licensed under Apache-2.0. Binary distributions
of the server are additionally governed by GPL-3.0 through the statically
linked nix-compat crate — see NOTICE for full attribution and
licensing details.

Full Changelog: v0.2.1...v0.3.0