fix(sfm): broaden Loc-N suffix regex to catch '.Loc' and 'Loc No.' variants

Operators use more separator variations than the original regex caught:
  - "Trumbull-Brayman-JV- Mont.Dam.Loc 2-R-25" — period as separator
  - "CMU - RKM Hall - Loc No. 3 - 4615 Forbes" — "No." between Loc and digit

Added period to the separator character class and optional "No." token
before the digit.  Catches both above patterns plus near-variants
without false-positives on normal project strings.

Real-data impact: 5 more clusters now auto-strip cleanly, including the
1,903-event Trumbull-Brayman-JV- Mont.Dam cluster.  Confidence
distribution: 43 → 44 high.

backend/services/metadata_backfill.py

@@ -104,15 +104,19 @@ def _normalise(s: Optional[str]) -> str:
 _PROJECT_LOC_SUFFIX = re.compile(
     r"""
     \s*                       # any leading whitespace
-    [-–—]             # hyphen or em-dash (separator before the Loc marker)
+    [-–—.]                    # separator: hyphen, em-dash, or period
-    \s*               # optional spaces
+                              # (operators use any of these — see
+                              #  "Mont.Dam.Loc 2-R-25")
+    \s*
     (?:loc|location)          # 'Loc' or 'Location'
-    \.?               # optional period
+    \.?                       # optional trailing period after Loc
-    \s*               # optional space
+    \s*
+    (?:no\.?\s*)?             # optional "No." or "No " before the digit
+                              # (e.g. "Loc No. 3", "Loc No 5")
     \#?                       # optional '#'
-    \s*               # optional space
+    \s*
     \d+                       # required digit
-    \b                # word boundary
+    \b
     """,
     re.IGNORECASE | re.VERBOSE,
 )