perf: PQ transform re-allocates a lot

[wjones127]

Using flat_map here breaks the size_hint() chain, preventing collect() from creating a perfectly sized allocation up front. Even though we should in theory know the exact size of the output.

lance/rust/lance-index/src/vector/pq.rs

Lines 120 to 136 in f30a679

	.flat_map(|vector| {
	vector
	.chunks_exact(sub_dim)
	.enumerate()
	.map(|(sub_idx, sub_vector)| {
	let centroids = get_sub_vector_centroids(
	codebook.values(),
	dim,
	num_bits,
	num_sub_vectors,
	sub_idx,
	);
	compute_partition(centroids, sub_vector, distance_type).map(|v| v as u8)
	})
	.collect::<Vec<_>>()
	})
	.collect::<Vec<_>>();

I noticed this while doing a benchmark with bytehound and sorting by number of allocations.

[wjones127]

These instructions for benchmarking are also relevant for this issue: #2838 (comment)