Fix incorrect unk_id assignment

`self.unk_id` is currently incorrectly assigned the ID of pad_token. Inferring from the original comment, the intention was likely to use `self.unk_token` (the second-to-last special token) to fetch its correct ID; this PR applies this correction.
However, I think this fix might affect model performance without retraining.

tokenization_kimia.py

@@ -127,7 +127,7 @@ class TikTokenTokenizer(PreTrainedTokenizer):
         self.pad_id: int = self.special_tokens[self.pad_token]
 
         self.unk_token: str = special_tokens[-2]
-        self.unk_id: int = self.special_tokens[self.pad_token]
+        self.unk_id: int = self.special_tokens[self.unk_token]
 
         self.stop_tokens = {
             self.special_tokens["[EOS]"],