Complete deriving diagnostics and fundep suite coverage #98

Part of #77 (Phase 4). The M5 derivation group, plus the class rows the corpus exercises that we have not covered.

Deriving

• CannotDeriveInvalidConstructorArg — 7 corpus cases, the largest single deriving group. Needs Generic/Rep support to determine whether a constructor's field type admits the derived representation.
• CannotDeriveNewtypeForData — 1 case.
• PossiblyInfiniteCoercibleInstance — 1 case.

FE-16 already covers role inference, Coercible solving, derive newtype adapters, and the Eq/Ord/Functor/Bifunctor/Contravariant derivings with Wasmtime evidence. What is missing is Generic-driven derivation for the remaining upstream classes and the diagnostics that say a derivation is impossible.

Also here

• Fundep ambiguity and consistency — FE-15 is Partial only on official-suite coverage. AmbiguousConstraint and FundepConflict need their purs counterparts chosen (Map TypeCheckErrorKind to official errorCodes #81 defers the mapping decision).
• Method-local class constraints — FE-14 records that scoped method-local constraints have rank-N evidence but no suite agreement.
• FE-15 remaining advanced class forms — multi-parameter classes are done; quantify what is left and land it.

Acceptance

The 7+1+1 deriving cases agree, FE-15's fundep rows have official-suite evidence, and FE-14/FE-15/FE-16 each state a concrete remaining gap rather than "reconcile official-suite coverage".