main-60be910

docs(perf): re-measure the regression sweep after the last two gates

The §19 sweep table stated 678 / 971 passing tests. Two gates have been added
since that measurement — `fdlimit`'s round trip through the platform's
`rlim_t` (P-12's portability defect) and `poll_file`'s content-wait (H-14) —
so the table under-counted its own tree.

Re-run on the final tree, not arithmetic:

  cargo test --release                              680 passed, 0 failed
  cargo test --release --features vpn,ssh-gateway   973 passed, 0 failed, 2 ignored
  cargo clippy --release --all-targets -- -D warnings          clean (both sets)

The count is in the table for the same reason the assertion count of
`public_idle_window.sh all` is: a matrix that silently loses — or gains — an
arm looks exactly like a matrix that passes.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BqfehFhjLR3Yj1sgaf5KCi