docs: the comment grep, what it found, and the first 7.0.0 entry

Records the sweep that found the macOS backend defect as a reusable command,
because it is cheap and it worked twice. A comment admitting the code cannot
determine something, sitting beside a returned value, is a reliable marker for a
fabricated reading.

Its second pass found one live defect and two latent ones, and the distinction
is easy to overstate: `ai_workload` reported any GPU-holding Python process as
PyTorch and that reached users through the tuner's evidence string; `dma_engine`
claimed a capability from an unreadable file and has no callers;
`cpufreq::is_turbo` encodes "cannot tell" as `false` and has no callers either.

The last one is the *Queued for the next major version* pattern recurring exactly
as that section predicted it would, so it opens the 7.0.0 list rather than being
fixed against a tag an hour old.

`src/audio/mod.rs` was checked and is clean -- its "placeholder" comment is in a
test, not a reader.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>