add Python interface

[Picus303]

This PR allows to use Fish Speech as an importable Python library for easy integration in code.
It adds the sub-directory lib in the main directory and mainly relies on the dedicated Pipeline class.

Here is an example of use:

# Optionnal: remove prints in terminal
import os
from loguru import logger

logger.remove()
os.environ["TQDM_DISABLE"] = "1"

# Prepare the models
from fish_speech.lib import Pipeline

model = Pipeline(
    llama_path = "models/fish-speech-1_5",
    vqgan_path = "models/vqgan-1_5.pth",
)

# Create a reference audio
ref = model.make_reference("ref.wav", "reference text")

# Generate audio (no streaming)
output = model.generate("text to generate.", ref)

# Generate audio (streaming)
import numpy as np

generator = model.generate("text to generate.", ref, streaming=True)

parts = []
for part in generator:
    parts.append(part)
    print(part.shape)

output = np.concatenate(parts, axis=0)

# Save the output to a file
sample_rate = model.sample_rate

import soundfile as sf
sf.write("output.wav", output, sample_rate)

This PR is not complete yet as it's missing:
1 - Documentation. Question: Where do you want to put it?
2 - The code still depends on .project-root to manage paths, making it impossible to install in non-editable mode, forcing the user to keep the source code in a separate folder. I'd be glad if you have suggestions for this part.