Regression tests for the pulled fixes (f59ada9)

- test_stitching_pad_mid_plane_close: solder-filled THT stitching pad
  mid-pour, adaptive vs uniform. Fails at 13.8% low on the pre-pinning
  adaptive path, passes at <0.1% with it.
- normalize_decimal / parse_frequency: decimal commas parse, thousands-
  separator patterns raise instead of silently scaling 1000x.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

tests/test_skin.py

@@ -64,6 +64,24 @@ def test_parse_frequency():
         skin.parse_frequency("-5k")
 
 
+def test_normalize_decimal():
+    """European decimal commas parse; thousands-separator patterns are
+    rejected ('1,500' silently becoming 1.5 was a 1000x input error)."""
+    assert skin.normalize_decimal("1,5") == "1.5"
+    assert skin.normalize_decimal("0,25") == "0.25"
+    assert skin.normalize_decimal("1,5000") == "1.5000"   # 4 digits: decimal
+    assert skin.normalize_decimal("2.5") == "2.5"
+    for bad in ("1,500", "1.500,5", "1,000,000", "12,345"):
+        with pytest.raises(ValueError, match="separator"):
+            skin.normalize_decimal(bad)
+
+
+def test_parse_frequency_decimal_comma():
+    assert skin.parse_frequency("1,5k") == 1500.0
+    with pytest.raises(ValueError):
+        skin.parse_frequency("1,500")     # ambiguous, not 1.5 Hz
+
+
 def test_single_layer_ac_scales_exactly():
     """Uniform conductance scaling leaves the field shape unchanged:
     R_AC = R_DC * factor to solver precision."""