feat(embeddings): expose token usage via embed_texts_with_usage

[sergiomeneses]

Summary

Add EmbeddingResponse struct and embed_texts_with_usage / embed_text_with_usage methods to the EmbeddingModel trait, enabling consumers to track token consumption when generating embeddings.

Currently embed_texts() returns only Vec<Embedding>, discarding provider-supplied token usage info.

Changes

• Trait: Add EmbeddingResponse struct (embeddings + usage) and embed_texts_with_usage() / embed_text_with_usage() to EmbeddingModel
• OpenAI: Override embed_texts_with_usage() with real usage from API response
• VoyageAI: Override embed_texts_with_usage() with real usage from API response
• EmbeddingsBuilder: Add build_with_usage() accumulating usage across all batches
• Backward-compatible: Existing embed_texts(), embed_text(), build() unchanged

Usage

// Direct with usage tracking
let response = model.embed_texts_with_usage(docs).await?;
println!("Tokens used: {}", response.usage.input_tokens);

// Via builder with usage tracking
let builder = EmbeddingsBuilder::new(model);
let (pairs, total_usage) = builder.build_with_usage().await?;

Related: #988
`

[sergiomeneses]

do you guys needs someting else to progress with this?