feat: add confidence-based RMP matching with manual review workflow

Replace simple auto-matching with scored candidate generation that
considers department overlap, name uniqueness, and rating volume.
Candidates above 0.85 auto-accept; others require admin approval.

migrations/20260130000000_collapse_instructors.sql

@@ -0,0 +1,80 @@
+-- Collapse instructors from per-banner-id rows to per-person rows (deduped by lowercased email).
+-- All existing RMP matches are deliberately dropped; the new auto-matcher will re-score from scratch.
+
+-- 1. Create the new instructors table (1 row per person, keyed by email)
+CREATE TABLE instructors_new (
+    id SERIAL PRIMARY KEY,
+    display_name VARCHAR NOT NULL,
+    email VARCHAR NOT NULL,
+    rmp_professor_id INTEGER UNIQUE REFERENCES rmp_professors(legacy_id),
+    rmp_match_status VARCHAR NOT NULL DEFAULT 'unmatched',
+    CONSTRAINT instructors_email_unique UNIQUE (email)
+);
+
+-- 2. Populate from existing data, deduplicating by lowercased email.
+--    For each email, pick the display_name from the row with the highest banner_id
+--    (deterministic tiebreaker). All rmp fields start fresh (NULL / 'unmatched').
+INSERT INTO instructors_new (display_name, email)
+SELECT DISTINCT ON (LOWER(email))
+    display_name,
+    LOWER(email)
+FROM instructors
+ORDER BY LOWER(email), banner_id DESC;
+
+-- 3. Create the new course_instructors table with integer FK and banner_id column
+CREATE TABLE course_instructors_new (
+    course_id INTEGER NOT NULL REFERENCES courses(id) ON DELETE CASCADE,
+    instructor_id INTEGER NOT NULL REFERENCES instructors_new(id) ON DELETE CASCADE,
+    banner_id VARCHAR NOT NULL,
+    is_primary BOOLEAN NOT NULL DEFAULT false,
+    PRIMARY KEY (course_id, instructor_id)
+);
+
+-- 4. Populate from old data, mapping old banner_id → new instructor id via lowercased email.
+--    Use DISTINCT ON to handle cases where multiple old banner_ids for the same person
+--    taught the same course (would cause duplicate (course_id, instructor_id) pairs).
+INSERT INTO course_instructors_new (course_id, instructor_id, banner_id, is_primary)
+SELECT DISTINCT ON (ci.course_id, inew.id)
+    ci.course_id,
+    inew.id,
+    ci.instructor_id,  -- old banner_id
+    ci.is_primary
+FROM course_instructors ci
+JOIN instructors iold ON iold.banner_id = ci.instructor_id
+JOIN instructors_new inew ON inew.email = LOWER(iold.email)
+ORDER BY ci.course_id, inew.id, ci.is_primary DESC;
+
+-- 5. Drop old tables (course_instructors first due to FK dependency)
+DROP TABLE course_instructors;
+DROP TABLE instructors;
+
+-- 6. Rename new tables into place
+ALTER TABLE instructors_new RENAME TO instructors;
+ALTER TABLE course_instructors_new RENAME TO course_instructors;
+
+-- 7. Rename constraints to match the final table names
+ALTER TABLE instructors RENAME CONSTRAINT instructors_new_pkey TO instructors_pkey;
+ALTER TABLE instructors RENAME CONSTRAINT instructors_new_rmp_professor_id_key TO instructors_rmp_professor_id_key;
+ALTER TABLE course_instructors RENAME CONSTRAINT course_instructors_new_pkey TO course_instructors_pkey;
+
+-- 8. Recreate indexes
+CREATE INDEX idx_course_instructors_instructor ON course_instructors (instructor_id);
+CREATE INDEX idx_instructors_rmp_status ON instructors (rmp_match_status);
+CREATE INDEX idx_instructors_email ON instructors (email);
+
+-- 9. Create rmp_match_candidates table
+CREATE TABLE rmp_match_candidates (
+    id SERIAL PRIMARY KEY,
+    instructor_id INTEGER NOT NULL REFERENCES instructors(id) ON DELETE CASCADE,
+    rmp_legacy_id INTEGER NOT NULL REFERENCES rmp_professors(legacy_id),
+    score REAL NOT NULL,
+    score_breakdown JSONB NOT NULL DEFAULT '{}',
+    status VARCHAR NOT NULL DEFAULT 'pending',
+    created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    resolved_at TIMESTAMPTZ,
+    resolved_by BIGINT REFERENCES users(discord_id),
+    CONSTRAINT uq_candidate_pair UNIQUE (instructor_id, rmp_legacy_id)
+);
+
+CREATE INDEX idx_match_candidates_instructor ON rmp_match_candidates (instructor_id);
+CREATE INDEX idx_match_candidates_status ON rmp_match_candidates (status);

migrations/20260131000000_instructor_rmp_links.sql

@@ -0,0 +1,24 @@
+-- Multi-RMP profile support: allow many RMP profiles per instructor.
+-- Each RMP profile still links to at most one instructor (rmp_legacy_id UNIQUE).
+
+-- 1. Create junction table
+CREATE TABLE instructor_rmp_links (
+    id SERIAL PRIMARY KEY,
+    instructor_id INTEGER NOT NULL REFERENCES instructors(id) ON DELETE CASCADE,
+    rmp_legacy_id INTEGER NOT NULL UNIQUE REFERENCES rmp_professors(legacy_id),
+    created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    created_by BIGINT REFERENCES users(discord_id),
+    source VARCHAR NOT NULL DEFAULT 'manual'  -- 'auto' | 'manual'
+);
+
+CREATE INDEX idx_instructor_rmp_links_instructor ON instructor_rmp_links (instructor_id);
+
+-- 2. Migrate existing matches
+INSERT INTO instructor_rmp_links (instructor_id, rmp_legacy_id, source)
+SELECT id, rmp_professor_id,
+       CASE rmp_match_status WHEN 'auto' THEN 'auto' ELSE 'manual' END
+FROM instructors
+WHERE rmp_professor_id IS NOT NULL;
+
+-- 3. Drop old column (and its unique constraint)
+ALTER TABLE instructors DROP COLUMN rmp_professor_id;