fix(core): preserve request fidelity and provider errors

[patlux]

Summary

• preserve nested JSON Schema, composition, constraints, and legacy OMP schema forms
• keep the legacy generate path explicitly text-only and reject images before network I/O
• normalize Command Code context overflow errors for Pi compaction/retry and redact provider error details
• cover stream, schema, image, overflow, and secret-redaction behavior

Dependency

Stacked on #37. Merge #37 first; this PR then retargets cleanly to main.

Verification

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

[github-actions]

Dependency Review

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

Scanned Files

None

[github-actions]

Memory benchmark

Compared base 83a228a with PR head ba14293 on the same GitHub-hosted darwin runner. Lower values are better.

Metric	Base	PR	PR − Base	Change
Stable RSS	169.0 MiB ± 0.4 MiB	167.8 MiB ± 0.3 MiB	-1.5 MiB ± 0.3 MiB	-0.9%
Physical footprint	111.4 MiB ± 0.5 MiB	111.1 MiB ± 0.6 MiB	-0.1 MiB ± 0.7 MiB	-0.1%
Physical peak	129.1 MiB ± 0.4 MiB	128.2 MiB ± 0.4 MiB	-1.1 MiB ± 0.5 MiB	-0.9%

Interpretation: Differences larger than their measured MAD: Stable RSS -1.5 MiB ± 0.3 MiB, Physical peak -1.1 MiB ± 0.5 MiB.

Extension overhead above pi baseline

Metric	Base overhead	PR overhead	Difference
Stable RSS	+8.2 MiB ± 0.8 MiB	+6.8 MiB ± 0.8 MiB	-1.5 MiB ± 0.3 MiB
Physical footprint	+5.1 MiB ± 0.5 MiB	+4.4 MiB ± 0.8 MiB	-0.1 MiB ± 0.7 MiB
Physical peak	+5.8 MiB ± 1.2 MiB	+5.1 MiB ± 0.6 MiB	-1.1 MiB ± 0.5 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.