The justfile header's measured counts drift, and nothing re-measures them

## What

The `justfile` header records seven measured numbers — three `--lib` totals and
four doctest totals — as the evidence for why no recipe passes `--all-features`
and why `loom-core` is outside `build_features`. It says to "treat the numbers
above as the reason and re-measure before widening either list" (#298).

Nothing re-measures them, and the work that moves them is not widening a list.
Measured on rustc 1.97.0, the toolchain the block itself names:

| command | recorded | measured |
| --- | --- | --- |
| `cargo test --lib` | 540 total, 3 `signal::` | **570**, 3 |
| `cargo test --lib --all-features` | 590 total, 0 `signal::` | **620**, 0 |
| `cargo test --lib --features build_features` | 585 total, 3 `signal::` | **615**, 3 |

Thirty high in each row, which is unit tests added since the block was written.
The four doctest rows were re-measured in #398, because two commits there moved
them; these three are untouched by it and are left as they are.

## Why this is filed rather than edited in passing

The differences the block argues from are intact. 620 − 615 = 5, which is the
"three `signal::` rows gained and eight lost" it names, and the `signal::`
column is unchanged in all three rows. So the reasoning is sound and only the
absolutes are stale — the case where retyping three numbers buys a record that
is stale again at the next test added.

Three shapes, and choosing between them is a judgment about what the block is
for rather than something the measurement settles:

- **Keep the absolutes and check them.** Something would have to run all three
commands and compare, which is a third feature set's worth of compilation for
a comment. Where such a check would live is the question #395 is filed about.
- **Record the differences instead of the totals.** The block exists to show
that `--all-features` removes `subscription::signal`'s tests and that
`build_features` keeps them; a row that states the delta and the `signal::`
count says that and does not move when an unrelated test is added.
- **Accept the drift** and say the numbers are a snapshot with a date, which is
what makes them reproducible without making them a promise.

## What is not in question

#298 is closed and its finding stands: the `signal::` columns still show what it
reported, and the doctest side now shows it more cleanly than when it was filed
(#398 removes a module's example that was collected under `loom-core` in a
configuration that could not compile it, so the `--doc` delta is exactly the two
`Signal` rows).

Raised while re-measuring the doctest rows for #398.

Start by reading the measured-count block in the justfile and the three cargo test commands it records. Review the alternatives in this issue and #395 to determine which representation maintainers want, then update the block accordingly and re-run the named commands to verify that the documented evidence is consistent.

由索引模型根据 Issue 内容生成。