Support returning multiple validation errors

Today, validation ends on the first validation error encountered.

This asks that all validation errors be aggregated in some kind of collection, so all validation errors (or at least, as many as possible or that makes sense) are returned.

This can be behind a generation flag so both behaviors can be generated. Generating both at once (so both ways are available at runtime) should probably not be a goal.

No file, test, or entry point is named. Start by locating the validation-generation path and its existing single-error behavior, then define the supported aggregation scope and generation-flag behavior before adding tests that distinguish the modes.

Written by the indexing model from the issue text.