Support for the QWen1.5-0.5B model

[chenghuaWang]

What's new?

• Support for the QWen1.5-0.5B model.
• A TikToken tokenizer example in qwen/tokenizer_qwen.hpp
• Sliding window attention's mask op

Bug fixed

• RMSNorm constructor bug involved in PR#75

[yirongjie]

Since we need to compile and run on Android using the NDK, you cannot use libraries like ICU and re2.

1. You need to run build_android.sh in the scripts to ensure it can be compiled.
2. If you need to use regular expressions, you can refer to the file Bpe.cpp, which includes some older APIs such as "#include <regex> \ #include <codecvt> "

[chenghuaWang]

The submodule has been removed in the latest commit.