(draft) tts: Sesame support

[pminev]

I've started a draft for sesame support. For now I only translated the safetensor models to gguf, so the next steps are to start implementing them. Not entirely sure if I've translated them 100% correctly, but will check them in the process and fix it.
I also had to add some properties to the safetensor config file:

{
    "architectures": [
        "LlamaForCausalLM"
    ],
    "n_layers": 128,
    "num_hidden_layers": 128,
    "n_layer": 128,
    "num_layers": 128,
    "num_attention_heads": 128,
^added
------------------------------------
    "backbone_flavor": "llama-1B",
    "decoder_flavor": "llama-100M",
    "text_vocab_size": 128256,
    "audio_vocab_size": 2051,
    "audio_num_codebooks": 32
}

Since they have 2 models in the model.safetensor I split them and then run convert_to_gguf

python3 split_hf.py

python3 convert_hf_to_gguf.py my-models/csm/backbone --outfile my-models/csm/gguf/csm-backbone-1B-f16.gguf --outtype f16
 
python3 convert_hf_to_gguf.py my-models/csm/decoder --outfile my-models/csm/gguf/csm-decoder-1B-f16.gguf --outtype f16

[pminev]

It's already implemented in #12648