refactor: introduce shared tokenizer abstraction and split implementations (#1423)

src/tokenizers/qwen2_tokenizer.h

@@ -0,0 +1,16 @@
+#ifndef __SD_TOKENIZERS_QWEN2_TOKENIZER_H__
+#define __SD_TOKENIZERS_QWEN2_TOKENIZER_H__
+
+#include <string>
+
+#include "bpe_tokenizer.h"
+
+class Qwen2Tokenizer : public BPETokenizer {
+protected:
+    void load_from_merges(const std::string& merges_utf8_str);
+
+public:
+    explicit Qwen2Tokenizer(const std::string& merges_utf8_str = "");
+};
+
+#endif  // __SD_TOKENIZERS_QWEN2_TOKENIZER_H__