[tisonkun]

Follow-up to #300 (comment).

Summary

• Introduce a private BucketCount to keep the bucket count and its precomputed remainder multiplier together.
• Move power-of-two masking and exact remainder reduction behind BucketCount::index.
• Preserve the public API, serialized bytes, and exact bucket assignment.

Trade-off

On the current 64-bit target, nesting the u32 count and u128 multiplier increases CountMinSketch from 96 to 112 bytes because the outer type cannot reuse the nested type's tail padding. Under main's existing inline-plus-heap convention, this also increases estimated_size() by 16 bytes per sketch. #302 independently makes estimated_size implementation-dependent; its current Count-Min implementation excludes this inline overhead.

Validation

• cargo x check
• cargo x test
• cargo x lint

[tisonkun]

I'll push a follow-up for change the estimated_size impl for mainly heap size.

[tisonkun]

Review after commit is welcome.

I'm glad to accept any better writing.