fix: preserve capabilities when model/1 receives a populated LLMDB.Model struct

[mhsdef]

Description

When ReqLLM.model/1 receives a fully-populated %LLMDB.Model{} struct (e.g. constructed via LLMDB.Model.new!/1 with explicit capabilities), it was tearing the struct apart, stripping nil fields, running it through LLMDB.Enrich.enrich_model/1, and rebuilding via LLMDB.Model.new(). This caused caller-provided fields like capabilities to be overwritten by Zoi schema defaults (e.g. json.schema: false, tools.enabled: false).

This matters because the recommended advanced usage path — constructing an explicit LLMDB.Model struct for models not yet in the catalog or with incorrect catalog data — was silently losing the capabilities the caller set. This caused determine_output_mode/2 to fall back to :tool_strict instead of using native :json_schema structured output, even when the caller explicitly declared json: %{schema: true}.

Root cause: LLMDB.Enrich.enrich_model/1 only derives family and provider_model_id — it doesn't do catalog lookups. After enrichment, the original non-nil values were being discarded. The fix merges enriched fields (family, provider_model_id) back under the caller's original values, so explicit capabilities are never overwritten.

Type of Change

• Bug fix (non-breaking change fixing an issue)
• New feature (non-breaking change adding functionality)
• Breaking change (fix or feature causing existing functionality to change)
• Documentation update

Breaking Changes

None. The existing behavior for bare structs (e.g. LLMDB.Model.new!(%{id: "foo", provider: :openai}) with no capabilities) is unchanged — enrichment still fills in family and provider_model_id.

Testing

• Tests pass (mix test) — 2814 tests, 0 failures
• Quality checks pass (mix quality)
• New test: verifies capabilities survive the model/1 round-trip for a fully-populated struct
• Existing enrichment test still passes (bare struct gets family/provider_model_id derived)

Checklist

• My code follows the project's style guidelines
• I have updated the documentation accordingly
• I have added tests that prove my fix/feature works
• All new and existing tests pass
• My commits follow conventional commit format
• I have NOT edited CHANGELOG.md (it is auto-generated by git_ops)

Related Issues

Discovered while investigating why determine_output_mode/2 was returning :tool_strict for models with json.schema: true capabilities — the capabilities were being silently dropped during model/1 normalization.

[mhsdef]

Nope.