fix(countsketch): restore hh_keys field + apply_delta topk-rebuild

[zzylol]

Summary

PR #36 silently reverted PR #37's hh_keys: Vec<String> field on CountSketchDelta (and the corresponding heavy-hitter rebuild branch in apply_delta) when its merge landed. PRs #38 (KLL seedable RNG) and #39 (module rename count -> countsketch) were branched off the post-#36 state, so neither restored the missing field. Result: main has been missing hh_keys ever since.

This PR restores the field on top of today's post-#39 module names (src/sketches/countsketch.rs), mirroring sketchlib-go's Delta.HHKeys heavy-hitter rebuild path. Unblocks ASAPQuery-backend's PR-#74 consumer to pin asap_sketchlib to main, which in turn unblocks ASAPCollector Phase 3 step 3 (backend consumes asap-precompute-rs).

Changes

• CountSketchDelta: add pub hh_keys: Vec<String> (additive; defaults via #[derive(Default)]).
• CountSketch: add pub topk: Vec<(String, f64)> with #[serde(default)] so legacy msgpack payloads still deserialize. Add topk_update helper that maintains a max-by-count heap bounded by COUNT_SKETCH_TOPK_CAPACITY = 100 (mirrors Go's TOPK_SIZE).
• CountSketch::apply_delta: after applying cell deltas, re-estimate every key in delta.hh_keys against the post-update matrix and feed it into self.topk via topk_update (mirrors Go's Delta.HHKeys loop in sketchlib-go/sketches/CountSketch/delta.go).
• New pub const COUNT_SKETCH_TOPK_CAPACITY re-exported from sketches::mod.

CountSketchDelta is a plain Rust struct (not proto-generated); no proto file regeneration was needed. The wire-format proto for CountSketchState already carries TopKState separately.

Out-of-range cell handling on apply_delta is left as fail-fast (existing main behavior); the silent-skip variant on refactor/wire-format-align-go is out of scope for this unblocker.

Test plan

• New unit test test_apply_delta_rebuilds_topk_from_hh_keys: builds a sketch with two known weighted keys, sends a delta carrying only hh_keys, asserts topk is rebuilt with re-estimated counts and that the heavier key out-ranks the lighter one.
• New unit test test_apply_delta_hh_keys_topk_capacity: sends COUNT_SKETCH_TOPK_CAPACITY + 5 keys via hh_keys and asserts the heap stays bounded at the cap.
• All existing tests updated to include hh_keys: vec![] in delta literals; 384 lib tests + 1 integration + 18 doctests pass.
• cargo fmt --all -- --check clean.
• cargo clippy --all-targets -- -D warnings clean.