card: declare per-locale templates / test configs so the viewer stops merging schemas

Delete 52 stale train_keywords.jsonl from v1 converter (MASSIVE has no keyword pool)

52 locales, per-row slots (one example per slot, no cross-utterance aggregation), README dataset card

Initial upload: 52 MASSIVE locales re-cast as intents-for-eval templates

initial commit