[SushantGautam]

Summary

Closes #61.

• Discover optional nested JSON model metadata on GET /v1/models/{id}, with a fallback to GET /v1/models (also recognizes Open WebUI's info.meta).
• Capture it exactly once per new run as frozen target metadata; record discovery status, source, and timestamp. Retries and historical views only read the frozen snapshot.
• Preserve arbitrary nested prompt bodies, versions, retrieval settings, tool definitions, and framework-specific fields, with a 256 KiB and depth limit plus secret-key redaction.
• Show one flattened, dot-path row per metadata key in /runs/<id>/ Frozen Manifest, revealing long values on expansion.
• Add per-key metadata comparisons to the Compare Runs Input Parameters table, including missing-key and changed-value highlighting.
• Document the custom compatibility extension and add tests for discovery, fallback, sanitization, freezing, and comparisons.

Validation

• Pure metadata discovery/flattening unit tests: 8 passing with isolated dependency shims.
• Source-edit anchors checked against current main and smoke-tested.
• Full Django test suite / GitHub CI: not run here (repo checkout and Django environment unavailable).

Compatibility

• Official OpenAI Model objects have no documented metadata field. This is optional and does not affect targets without the extension.
• No database migration; existing target snapshot JSON field is reused.