Add Embedding Related Functionality

[Hirtol]

Recently llama.cpp has added the ability to generate embeddings using BERT related models. There have been some issues, namely differences between the embeddings generated in Python and the ones coming from llama.cpp, but with ggerganov/llama.cpp#5796 being merged it seems to finally be stable.

This PR implements the functionality to allow a user to make use of these new features, it adds an example based loosely on the llama.cpp embeddings example. Most of the code was translated from either llama.cpp or llama-cpp-python (specifically, the add_sequence method in LLamaBatch).

A few additional changes snuck in as well, replacing a println! call with a tracing call to allow users to filter it out, alongside a fix for a failing doc-test.

[MarcusDunn]

this looks good. one question about consolidating some unsafe.

[Hirtol]

Hm, the doc-test that I 'fixed' fails here. llama_token_type seems to get translated to c_uint on CI, but c_int for me locally.

EDIT: Ah, seems to be a Windows vs. Linux issue: rust-lang/rust-bindgen#1966
(as a side-note: pprof currently prevents the running of tests on Windows by default unless manually commented out)

[MarcusDunn]

fix that and I'll add windows to the test CI so I don't break it again. (as llama_token_type where needed should work fine)

[MarcusDunn]

thanks for the quick turnaround. Just cut a new release.