Change add_bos_token to true

Otherwise this causes the special tokens in chat_template to not be applied correctly in certain inference frameworks, such as llama.cpp.

tokenizer_config.json

@@ -1,5 +1,5 @@
 {
-  "add_bos_token": false,
+  "add_bos_token": true,
   "add_eos_token": false,
   "add_prefix_space": true,
   "added_tokens_decoder": {