bench(cli): measure cold executable startup and verify diagnostics

[nth-bailey]

Summary

• Add benchmarks/cli/startup.py, a Linux fresh-process benchmark that alternates warm runs with runs after requesting eviction of the CLI executable's file pages. It records per-run time and major page faults in JSON.
• Document the method and its limits alongside the existing warm-cache hyperfine benchmark.
• Verify that misspelled backends, incompatible backends, and invalid features produce actionable messages; clap colors them in a TTY and omits ANSI escapes in redirected output or under NO_COLOR=1. No CLI diagnostic code change was needed.

Result

On this WSL2/ext4 host, 100 release-build runs per case and mode measured median --help times of 1.52 ms warm and 7.61 ms after executable-page eviction; invalid-backend times were 1.64 ms and 7.71 ms. All 100 evicted runs in each case had major page faults. This is a controlled cold-executable proxy, not a post-reboot or cross-platform measurement. The <5 ms figure in #50 was an aspirational proposal, not a release requirement; no startup defect was found. #63 was closed as not planned after that clarification.

Verification

• Release build under scripts/memcap.sh
• python3 benchmarks/cli/startup.py --runs 100 and a formatted-script smoke run
• PTY and redirected-output checks for three invalid option cases, with and without NO_COLOR
• Ruff check/format, git diff --check, pre-commit hooks, pre-push workspace cargo check

Closes #62. #63 was closed as not planned.