feat(models): add model-aware runtime metadata

[patlux]

Summary

• enrich only known models with supported reasoning levels and forward mapped reasoning_effort
• bound model discovery with a configurable timeout and validated cache fallback
• add /commandcode-refresh and /commandcode-status with coalescing and redacted diagnostics

Verification

• npm test
• npm run format:check
• git diff --check

Notes

• OMP process test skips because omp is not installed on the test host; legacy nested reasoning metadata remains covered by unit tests.

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

[github-actions]

Memory benchmark

Compared base 8562009 with PR head c0d126c on the same GitHub-hosted darwin runner. Lower values are better.

Metric	Base	PR	PR − Base	Change
Stable RSS	167.0 MiB ± 0.4 MiB	168.1 MiB ± 0.5 MiB	+1.3 MiB ± 0.9 MiB	+0.8%
Physical footprint	110.4 MiB ± 0.4 MiB	111.0 MiB ± 0.8 MiB	+0.9 MiB ± 0.4 MiB	+0.8%
Physical peak	126.6 MiB ± 0.5 MiB	127.6 MiB ± 0.6 MiB	+1.7 MiB ± 1.3 MiB	+1.3%

Interpretation: Differences larger than their measured MAD: Stable RSS +1.3 MiB ± 0.9 MiB, Physical footprint +0.9 MiB ± 0.4 MiB, Physical peak +1.7 MiB ± 1.3 MiB.

Extension overhead above pi baseline

Metric	Base overhead	PR overhead	Difference
Stable RSS	+6.6 MiB ± 0.6 MiB	+7.5 MiB ± 1.6 MiB	+1.3 MiB ± 0.9 MiB
Physical footprint	+4.0 MiB ± 0.3 MiB	+4.6 MiB ± 0.5 MiB	+0.9 MiB ± 0.4 MiB
Physical peak	+3.2 MiB ± 0.6 MiB	+4.6 MiB ± 0.8 MiB	+1.7 MiB ± 1.3 MiB

Values are medians of 6 alternating, paired runs. The value after ± is the median absolute deviation (MAD). Each process was sampled 12 times after a 2500 ms warm-up.

Environment: pi 0.82.1, Bun 1.3.11, darwin arm64. RSS comes from ps; physical footprint and peak come from macOS footprint.

This is a comparative signal, not a pass/fail threshold. GitHub-hosted runner noise can affect absolute values.

[patlux]

Synced reasoning metadata exactly with the current published command-code@1.14.1 model catalog:

• removed five models whose CLI catalog has no adjustable efforts
• added Qwen/Qwen3.8-Max with low, medium, xhigh
• removed the inferred defaultLevel
• separated Pi's nullable thinkingLevelMap from canonical OMP thinking: { mode: "effort", efforts, effortMap }
• added an exact catalog regression fixture

Verified with npm test, npm run format:check, and git diff --check.