DimensionsMetadata models lack schema validation; malformed dimensions.json fails late

[matt-edmondson]

Summary

The deserialised metadata classes in Semantics.SourceGenerators/Models/ (e.g. DimensionsMetadata) accept any JSON shape. There is no post-deserialisation validation for required fields (Name, Symbol, base unit, supported forms), unit-string non-emptiness, or referential integrity across dotProducts / crossProducts / integrals / derivatives.

A malformed entry does not surface until the generator either crashes mid-emit or (more commonly, see #56) silently drops the operator.

Suggested next step

• Add a Validate() method on DimensionsMetadata that runs after JSON deserialisation and throws (or reports diagnostics) for: missing Name/Symbol, empty availableUnits, undeclared form references, unknown referenced dimensions.
• Run it once at the top of each generator pass and bail out with a clear diagnostic when validation fails.

Goes hand in hand with #56 (diagnostics for unresolved references).

Area / Severity

SourceGenerators / Metadata · cleanup

[matt-edmondson]

Closed by #70 — DimensionsMetadata.Validate() now runs at the top of QuantitiesGenerator and reports SEM002 for missing name/symbol, empty availableUnits, blank unit entries, dimensions that declare no vector forms, missing base on a form, and duplicate base/overload type names. Cross-dimension reference checks remain in SEM001.

---

Generated by Claude Code