Implement generate_bufdir_report RPC Function

Structure the function as a series of CTEs for readability: (1) `filtered_activities` — activities in org + date range, (2) `mapped_activities` — JOIN with bufdir_category_mappings on internal_type_id + mapping_version, (3) `aggregated` — GROUP BY bufdir_code with COUNT DISTINCT participant_id and SUM hours, (4) `final_result` — format as JSONB array. For deduplication of proxy vs direct: use a UNION approach where you first collect all (participant_id, bufdir_code) pairs regardless of is_proxy_registered, then COUNT DISTINCT — this naturally deduplicates. The is_proxy_registered flag is only relevant if you want to report separately on proxy vs direct hours (keep this as a future extension but implement COUNT DISTINCT naively first). Return JSONB with `json_agg(json_build_object(...))` for the result array.

Handle the NULL case (no activities) by returning an empty JSON array `'[]'::jsonb` not NULL.

SQL unit tests using pgTAP or manual test scripts: (1) insert known test activities and verify COUNT DISTINCT matches expected unique participant count, (2) verify proxy-registered activities do not inflate participant_count for participants who also have direct registrations, (3) verify activities outside the date range are excluded, (4) verify unmapped activity types appear in 'unmapped' category, (5) verify a user from org B cannot retrieve org A's data by passing org A's p_org_id. Performance test: load 50,000 rows for a test org and EXPLAIN ANALYZE the function to confirm index usage. Test with p_mapping_version that has no entries — function must return empty array, not error.