Bump llama_tokenize APIs to latest specs

[manishshettym]

The llama_tokenize and llama_tokenize_with_model APIs are out-dated with respect to the equivalent APIs in llama.cpp (as shown below):

1. https://github.com/ggerganov/llama.cpp/blob/7ddf185537b712ea0ccbc5f222ee92bed654914e/llama.h#L374
2. https://github.com/ggerganov/llama.cpp/blob/7ddf185537b712ea0ccbc5f222ee92bed654914e/llama.h#L382

ISSUE: Particularly, they are missing an argument text_len which denotes the length of the text/prompt being tokenized. This causes segmentation faults, as the wrong arguments get passed on to the C++ library.

This PR adds the required arguments to llama_cpp.py. cc: @abetlen

[manishshettym]

Ref to the relevant change in llama.cpp's API (two days ago): ggml-org/llama.cpp#3170

[manishshettym]

closing this PR since the changes were made directly by @abetlen! Thank you.