Match Qwen Unicode number token boundaries

docs/issue111-tokenizer/normalization-gap.json

@@ -0,0 +1,38 @@
+[
+  {
+    "text": "́a á ́ १́b",
+    "ids": [
+      52033,
+      64,
+      264,
+      52033,
+      220,
+      52033,
+      220,
+      12256,
+      100,
+      52033,
+      65
+    ],
+    "full_ids": [
+      52033,
+      64,
+      18135,
+      220,
+      52033,
+      220,
+      12256,
+      100,
+      52033,
+      65
+    ],
+    "pieces": [
+      "́a",
+      " á",
+      " ́",
+      " ",
+      "१",
+      "́b"
+    ]
+  }
+]