Add Hugging Face Transformers.js support for local embeddings

[yencolon]

Adds support for local, offline embeddings using Hugging Face's Transformers.js library. This enables Vectra to generate embeddings without requiring API calls to OpenAI or Azure, making it ideal for privacy-sensitive applications, offline environments, and reducing operational costs.

Fixes #96

Key additions:

• TransformersEmbeddings: New embeddings model supporting popular models like Xenova/all-MiniLM-L6-v2, Xenova/bge-small-en-v1.5, and Xenova/bge-base-en-v1.5
• TransformersTokenizer: Tokenizer that aligns with Transformers.js models for accurate text chunking
• Optional dependency: @huggingface/transformers added as an optional peer dependency with dynamic imports.

Example Usage

const embeddings = await TransformersEmbeddings.create({
    model: 'Xenova/all-MiniLM-L6-v2',
    device: 'auto',
    dtype: 'q8'
});

// Use with LocalDocumentIndex
const index = new LocalDocumentIndex({
    folderPath: './my-index',
    embeddings: embeddings,
    tokenizer: embeddings.getTokenizer()
});