[SizzleUnrlsd]

Item 3 of #21. Stacked on #23 (test/fixture-expectations): merge #22 and #23 first.

Change

test/cli/check_cli.py (the cli test) proves every option as the README's table states it, on small programs written into the work directory (not under test/, which the sweeps run in full):

• --run-arg values reach the program as arguments, in order; --env NAME=VALUE sets the variable, and nothing leaks into the program without it; --cwd runs the program in that directory, with a relative -o too.
• --show-output prints the program's stdout and stderr after the summary, --show-events the CoreTrace event lines; neither is printed without the option.
• --format text and --format sarif as a separate argument, --output=<path> with an equals sign, --help on stdout with exit 0.
• Batch mode: a program exiting non-zero is [RUNTIME] on its line and counted in runtime_failures, but is not a finding (exit 0); --strict-test-exit raises the status to 1; --output-dir places the binaries; --test-dir is repeatable; a directory that does not exist gives exit 2 and a message.
• Every ParseArgs error path exits 2 with its message and the usage on stderr: unknown option, -o/--run-arg/--env/--cwd/--timeout/--test-dir/--output-dir/--format without a value, --env without NAME= or with an empty name, a negative --timeout, no compiler arguments with or without --. -o given both to the analyzer and to the compiler is refused too.

README: the cli row of the tests table.

Tests (written first)

The checks are the deliverable, written from the README's option table before running them; all 48 passed on the first run, so the documented behaviour holds. To prove they bite, I removed the handlers of --run-arg, --env, --cwd, --strict-test-exit, --show-output and --show-events and the unknown-option message: nine checks fail, one or two per option, and none of the others.

Verification

• macOS arm64: ctest --test-dir build 14/14 (cli 8 s); ./scripts/format-check.sh clean.
• Linux: this pull request's Build workflow.