fix: prevent overflow at kmeans

[cutecutecat]

Thanks to @xieydd , we have digged out this bug.

Panickied. Info: PanicInfo { payload: Any { .. }, 
message: Some(attempt to subtract with overflow), 
location: Location { file: "crates/service/src/algorithms/clustering/[elkan_k_means.rs](http://elkan_k_means.rs/)", line: 179, col: 47 }, 
can_unwind: true, force_no_backtrace: false }

If there is too few vectors with a small count n and larger nlist as c, it will be a overflow there.

[usamoi]

should we just return centroids using a simpler algorithm if n <= c?

[cutecutecat]

should we just return centroids using a simpler algorithm if n <= c?

Fine, I will try QuickCenters by https://github.com/pgvector/pgvector/blob/cc9e6a67783092ae2e8b1b1ade552e547bef8934/src/ivfkmeans.c#L117

Different from pgvector implement, we don't exclude duplicates.

[usamoi]

QuickCenters is already in the code.