AI-grade an implementation against the intent spec. Evaluates each outcome, constraint, constitution rule, and edge case via Gemini structured output; stores the result on the intent and logs a verification_checkpoint implementation note.

category enum is outcome|constraint|constitution|edge_case; status enum is pass|fail|unclear. pass is true only when zero items fail; unclear counts as half credit toward score (0-100). Side effects: writes verification_result + updated_at on intent_specs, and inserts an intent_implementation_notes row with source 'verification_checkpoint'. No rate limiter on this route despite being AI-backed.