embed_batch does not use parallel computation

[EricLBuehler]

Hi @ShelbyJenkins!

I was taking a look at this great crate, and I noticed that embed_batch does not use batching.

candle_embed/src/lib.rs

Lines 261 to 273 in 3036c6f

	pub fn embed_batch(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>> {
	if texts.is_empty() {
	return Err(Error::msg(
	"CandleEmbed error: embed_batch called with empty texts",
	));
	}
	let mut embeddings = vec![];
	for text in texts {
	let embedding = self.embed_one(text)?;
	embeddings.push(embedding);
	}
	Ok(embeddings)
	}

Perhaps you can add this feature by padding the texts and then unpadding the embeddings? It is just a thought, but I think it would improve performance significantly for real-world use cases!