fix: don't hand nextest a phantom-only filterset in run

[cargo-affected-bot]

Deleting a test makes cargo affected run exit 4 with error: no tests to run.

Deleting a test function produces a diff hunk over exactly the lines its own stored coverage range covers, so the deleted test is selected as affected — and if the deletion is the only change, it is the whole selection. It is also gone from the current nextest list, so the generated default-filter matches nothing, and nextest's "no tests to run" exit propagates through run_tests. A stale cache row is reported to the caller as a test failure.

collect --diff has never had this problem: it passes --no-tests=warn and uses DiffPlan::live_selected_count afterwards to tell an all-phantom rerun from a runner-shim failure. run had no equivalent.

Repro

Against the two-module scratch project the functional suite uses, with the whole test_multiply block deleted and nothing else changed:

1 tests to run (1 affected + 0 config + 0 new + 0 stranded, 1 skipped of 2 reachable-known)
running 1 tests with nextest
    Starting 0 tests across 1 binary (1 test skipped via profile.default.default-filter)
     Summary [   0.000s] 0 tests run: 0 passed, 1 skipped
error: no tests to run

Exit code 4.

Change

Selection::live_selected() (new, in selection.rs) is selected() restricted to tests still in the listing. run hands nextest that subset instead of the full selection, and short-circuits with exit 0 when it is empty — there is no run to make, and asking nextest to run a phantom-only filterset can only produce the exit above. Both paths emit a note naming the count and pointing at cargo affected collect to drop the rows.

status gets the same note, in the conditional tense. Without it the dry run would predict a test that run now visibly skips, which is the invariant #71 established.

collect --diff is untouched — it keeps phantoms in its filterset deliberately, and prunes their rows afterwards.

Tests

run_with_all_phantom_selection_exits_zero (functional) collects, deletes test_multiply, and asserts exit 0 plus the two new messages. Verified failing against pre-fix src/: run with an all-phantom selection must exit 0, got Some(4), with error: no tests to run in the captured output.

Two unit tests in selection.rs pin live_selected()'s filtering and the notice's singular/plural agreement.

cargo test is green (114 unit + 42 functional); cargo clippy --all-targets is clean.

Relationship to #39 / #41

Same exit-4 failure class, different trigger, disjoint fix. #41 keeps filter-mismatched tests (positional filters, -E, #[ignore]) out of the selection by keying off filter-match.status; its own comment notes that phantoms are deliberately not in listing.excluded and stay in affected. This handles that remaining source. The two should merge cleanly enough — #41 edits the top of compute and the imports in run.rs; this edits impl Selection and the tail of run().