Validator walks PlainInputArg.fields() for UnboundField rejection

R215’s validator-side check on InputField.UnboundField (per the spec’s §Validator rules: condition.isPresent() && !condition.get().override() → reject at the directive’s source location) only walks GraphitronType.TableInputType.inputFields() via GraphitronSchemaValidator.validateTableInputType + validateInputFieldRecursive. Truly plain input types (GraphitronType.InputType permits ; JavaRecordInputType, PojoInputType, JooqRecordInputType, JooqTableRecordInputType) carry no classified InputField records on the type; their fields classify at consumer time on ArgumentRef.InputTypeArg.PlainInputArg.fields(). The validator’s whole-schema walk has no view into PlainInputArg, so plain-input UnboundField + @condition(override:false) shapes escape the validator-mirrors-classifier rule. R215’s acceptance test #5 (r215_validatorRejectsOverrideFalseOnNonBindingField) was written against a @table input (which routes through the existing walker) so the test passes, but the literal spec phrasing "non-binding plain input field" is structurally unreachable today.

The symptom that surfaces this: a plain input field with @condition(override:false) inside an @condition(override:true) cascade silently fires the explicit @condition method (the consumer-arm patch shipped in R215’s late round restored the doc-promised "every @condition you write produces SQL" contract for the cascade case) but the no-column shape is still malformed and the developer doesn’t get a build-time error. The validator’s job is to catch that regardless of where the field lives.

Implementation: walk PlainInputArg.fields() from each QueryField / MutationField at validate time, reusing GraphitronSchemaValidator.validateInputFieldRecursive (already recurses through InputField.NestingField). The walk surface is reachable via SqlGeneratingField.argumentRefs() or the equivalent per-variant accessor; the per-arg switch picks PlainInputArg and feeds its fields through the same recursion the @table walker uses. No new validator method required; just a second entry point that fans out by argument shape rather than by type shape.

Surfaced by the R215 self-review: alf flagged that the architect’s brief glossed the cascade contract as "outer override:true drops inner @condition" when the documented rewrite reading is "outer override:true suppresses only implicit predicates; inner @condition annotations always fire." The shipped consumer-arm patch closed the doc-contract bug; this item closes the validator gap that the consumer-arm patch was implicitly relying on the validator to cover.