[MirzaMerdovic]

Summary

This change ports the checker tests of the bats suite to Rust integration tests in
crates/check/tests, and removes the bats suite from cargo test. The base branch is
rust-port (#8).

Rust file	bats file	Tests
fail.rs	fail.bats	23
pass.rs	pass.bats	17
fixtures.rs	fixtures.bats	2
cli.rs	cli.bats	7
config.rs	config.bats	15
suppress.rs	suppress.bats	4

• Each test runs the checker as a separate process. CHECK selects the checker, as in the
  bats suite, so the same tests also run the bash checker.
• tests/common/mod.rs holds the helpers. Repo is a temporary git repository from the
  tempfile crate. tempfile is the only new dependency, and only the tests use it.
• A fail test compares stdout with the .expected file byte for byte and requires an
  empty stderr. The bats test compares stdout and stderr together.

Not changed

• The bats suite, tests/run, and tests/mutate are unchanged. tests/mutate still uses
  the bats suite for the bash checker.
• repo.bats and snippets.bats test repository content, not the checker. They stay in
  bats.
• The Rust workflow still runs tests/run against the release binary.

Verification

• The 68 Rust integration tests pass against the Rust binary and against the bash checker.
• tests/run passes 73 of 73 tests with both checkers.
• cargo mutants --jobs 32 found 194 mutants: 185 caught, 9 unviable, no missed mutants,
  and no timeouts.

Mutation run time

These local runs used 32 jobs. Another process used about 5 processors during the runs,
so the times are approximate.

Tests for each mutant	Wall-clock time	Mean test time per mutant
Unit tests and the bats suite	135.8 s	11.0 s
Unit tests and the Rust integration tests	85.3 s	0.7 s

On 4 processors in CI, the full mutation run took 317 s on #8 and 228 s on the previous
commit of this change.

Mutation check scope

A pull request now runs cargo mutants --in-diff on the diff between its merge commit and
the base branch. A push to main runs every mutant.

• A pull request that changes no source line tests no mutants. This pull request is one.
• A local run on a diff that changes one line tested 5 mutants with 4 jobs in 6 s.
• A pull request that changes only tests can remove coverage without a failed check. The
  full run on main then reports the missed mutants.

Debug build

cargo test runs the debug build of the binary. Without optimization, the regex crate
compiles the rule expressions slowly. One run on idiom.md took 139 ms with the debug
build and 10 ms with the release build.

An optimized dev profile for the dependencies did not help cargo mutants. Each of its
build directories then compiles regex with optimization, and a local run took 229 s
instead of 136 s. This change keeps the default profile.