feat: persist and load explained variance for scree plots

- compute_svd_for_window now computes explained variance ratio (s²/sum(s²))
  and appends it as a metadata row (entity_type='metadata',
  entity_id='explained_variance') to motion_rows
- load_scree_data reads this metadata row from svd_vectors instead of
  querying the non-existent sv_metadata column
- run_svd_for_window counts only entity_type='motion' rows in stored_motion
  so metadata rows don't inflate the count
- Added 5 TDD tests covering load, compute, store, and round-trip

All 227 tests pass.

pipeline/svd_pipeline.py

@@ -409,11 +409,16 @@ def compute_svd_for_window(
             for j, mid in enumerate(motion_ids)
         ]
 
+        # Persist explained variance ratio as a metadata row for scree plots
+        evr = (s ** 2 / np.sum(s ** 2)).tolist()
+        motion_rows.append(("metadata", "explained_variance", evr, None))
+
         return {
             "window_id": window_id,
             "k_used": k_used,
             "mp_rows": mp_rows,
             "motion_rows": motion_rows,
+            "explained_variance": evr,
         }
 
     except Exception:
@@ -438,8 +443,10 @@ def run_svd_for_window(
 
     rows = result["mp_rows"] + result["motion_rows"]
     stored = db.batch_store_svd_vectors(window_id, rows)
+    # motion_rows may include metadata rows (e.g. explained_variance)
+    motion_entity_rows = [r for r in result["motion_rows"] if r[0] == "motion"]
     return {
         "k_used": result["k_used"],
         "stored_mp": len(result["mp_rows"]),
-        "stored_motion": len(result["motion_rows"]),
+        "stored_motion": len(motion_entity_rows),
     }