Support Stella-EN 1.5B v5 pre-tokenizer

[shermanhuman]

Currently llama.cpp doesn't seem to support Stella-en-1.5b-v5:

Error logs from LM Studio:

2025-01-09 09:40:48 [DEBUG]
llama_model_load: error loading model: error loading model vocabulary: unknown pre-tokenizer type: 'stella_en_1.5B_v5'
llama_load_model_from_file: failed to load model
2025-01-09 09:40:48 [DEBUG] llama_init_from_gpt_params: failed to load model 'C:\Users\s\.cache\lm-studio\models\Losspost\stella_en_1.5b_v5\losspost-stella-en-1.5b-v5.gguf'
2025-01-09 09:40:48 [DEBUG] lmstudio-llama-cpp: failed to load model. Error: llama.cpp error: 'error loading model vocabulary: unknown pre-tokenizer type: 'stella_en_1.5B_v5''

This Qwen based model scores highly on the MTEB leaderboard. I think running embedding models locally are a big win for privacy, and so a great addition to llama.cpp.

It does run on Ollama now, so maybe we can borrow from their fork?