fix(exec): align exec help usage with parser

[middleDuckAi]

Summary

• align codex exec --help with the actual parser contract
• advertise the two real exec invocation forms instead of a phantom second positional
• add a regression test for the exec help output

Why

Issue openai#15535 reports that codex exec --help advertises Usage: codex exec [OPTIONS] [PROMPT] [COMMAND], but the parser rejects a second positional argument. That makes the help output misleading for both humans and automation.

This change keeps parser behavior unchanged and fixes the contract shown in help instead:

• codex exec [OPTIONS] [PROMPT]
• codex exec [OPTIONS] <COMMAND> [ARGS]

Testing

• cargo fmt --all --check
• cargo run -p codex-cli -- exec --help | sed -n '1,20p'
• cargo test -p codex-cli exec_help_does_not_advertise_a_second_positional
• cargo test -p codex-cli exec_resume
• printf '' | cargo run -p codex-cli -- exec --ephemeral --skip-git-repo-check --color never - false

Notes

• Parser behavior is intentionally unchanged.
• This PR only fixes the help/usage contract so automation and users are not misled by [PROMPT] [COMMAND].