feat(cli): --list / --dry-run to show which tests would run - #1034

Every selection mechanism — --filter, --tag, --exclude-tag, --shard,
--random-order --seed, --rerun-failed, file::fn, file:LINE — was only
observable by actually running the suite. Checking shard balance took one
full run per shard, and a typo'd --filter was indistinguishable from a
genuinely empty suite.

--list prints `path::function` per line on stdout with the count on stderr,
so it pipes cleanly; --list-format json adds name, line and tags.
--dry-run is a shellspec-compatible alias.

The listing hooks into load_test_files after every selection step and
before render_running_file_header and set_up_before_script, so nothing runs
beyond sourcing the files. Ordering is shared with the runner via the
extracted runner::order_functions_for_script, so a listed order cannot
drift from the order that seed executes.

An empty selection exits 0: this is a query, not a run.

Closes #1007

…red output

The seed-order test parsed result lines with `sed 's/\x1b.../'` and
`grep -oE '^. Passed:'`. Both are environment-dependent — \x1b is a GNU sed
extension, and the status glyph is multi-byte, so `.` matches a different
amount under a non-UTF-8 locale. It passed on macOS and failed on every
Linux and locale job.

The fixture now records its own execution order to a file, so the test
compares orders directly with no rendering in the path. A paired test
asserts the seed actually reorders, so the comparison cannot hold trivially.