[CLI] Measure cold startup and verify interactive error diagnostics #62

Context

#50 asks for a cold polyxml --help startup target below 5 ms and clear, colorized errors for invalid backend/style/feature combinations. benchmarks/cli/benchmark.sh already measures 100 fresh processes with warm OS caches (about 1.4–1.6 ms on the documented host); its README explicitly says this does not establish cold-disk or post-reboot latency. The CLI's target-validation errors are actionable plain-text InvalidInput errors; terminal color behavior has not been verified.

Scope

• Define a reproducible meaning of cold startup (for example, first process after a fresh VM boot or a controlled file-cache state) and document the host and measurement method. Keep the existing warm-process results as a separate metric.
• Measure top-level help and argument validation with hyperfine or an equivalent reliable timer. Report measured latency and its limits; the original 5 ms proposal is not a release requirement.
• Verify error readability for misspelled backends and invalid target combinations in an interactive terminal and with redirected output. Add color only where it improves the interactive output; avoid escape codes in redirected output.
• Add focused CLI tests for any changed diagnostic behavior and update benchmarks/cli/README.md with the results and limitations.

Related: #50, PR #56.

Resolution

PR #64 measured warm and executable-evicted startup and verified terminal, redirected, and NO_COLOR diagnostics. No CLI error-handling bug was found. The 5 ms figure in #50 was an unsupported aspirational proposal and is not being retained as a requirement. The results are documented in benchmarks/cli/README.md.