schema: add composite index (is_published, year DESC) + fix stale migration 005

- Add idx_theses_pub_year composite index on theses(is_published, year DESC) to
  schema.sql; replaces the need for the query planner to pick between the two
  separate idx_theses_published / idx_theses_year indexes and sort with a temp
  B-tree. Every public query filters on is_published=1 and orders/filters by year,
  so this covering index eliminates the sort pass for those queries.

- Create storage/migrations/006_add_composite_index.sql and apply to both
  posterg.db and test.db.

- Fix storage/migrations/005_add_banner.sql: the view recreation in that file
  still referenced the pre-migration-001 table/column names (thesis_keywords,
  keywords.keyword). Updated to use thesis_tags / tags tg to match the canonical
  schema.sql. The live DB was unaffected (migration 001 ran before 005), but the
  file was misleading and would fail if ever re-run from scratch.

storage/migrations/005_add_banner.sql

@@ -38,7 +38,7 @@ SELECT
     GROUP_CONCAT(DISTINCT CASE WHEN ts.role = 'lecteur'    THEN s.name END) as jury_lecteurs,
     GROUP_CONCAT(DISTINCT l.name) as languages,
     GROUP_CONCAT(DISTINCT fmt.name) as formats,
-    GROUP_CONCAT(DISTINCT k.keyword) as keywords
+    GROUP_CONCAT(DISTINCT tg.name) as keywords
 FROM theses t
 LEFT JOIN orientations o ON t.orientation_id = o.id
 LEFT JOIN ap_programs ap ON t.ap_program_id = ap.id
@@ -53,8 +53,8 @@ LEFT JOIN thesis_languages tl ON t.id = tl.thesis_id
 LEFT JOIN languages l ON tl.language_id = l.id
 LEFT JOIN thesis_formats tf ON t.id = tf.thesis_id
 LEFT JOIN format_types fmt ON tf.format_id = fmt.id
-LEFT JOIN thesis_keywords tk ON t.id = tk.thesis_id
-LEFT JOIN keywords k ON tk.keyword_id = k.id
+LEFT JOIN thesis_tags tt ON t.id = tt.thesis_id
+LEFT JOIN tags tg ON tt.tag_id = tg.id
 GROUP BY t.id;
 
 -- Recreate public view