feat: add frequencies sketches

[PsiACE]

• support FrequentItemsSketch
• add baisc test cases and serde tests

[PsiACE]

only 1 serde case failed on windows, ... on windows, long is 32‑bit......

[jmalkin]

Probably don’t need to bother with a dedicated implementation for longs. That was mostly for initial testing, I think. And if rust supports templates/generics with primitives then there’s no reason for a separate implementation for longs. We didn’t do one for c++, just provided a default serde for a few basic types.

[PsiACE]

you're right.

just provided a default serde for a few basic types.

I'll mark this pr as a draft for now. Later, I'll remove longsketch, update the tests, and skip the serde errors on Windows.

[leerho]

I need to see some Rust reviewers before I approve & merge this.

[tisonkun]

I'll take a look here before the new year :P

[tisonkun]

Rest LGTM. Mostly style comments to keep our codebase consistent.

[tisonkun]

@leerho There are several further potential improvements. But those can be follow-ups.

I'd leave it to @PsiACE to decide whether to include it in this patch or in a follow-up PR.

[tisonkun]

@PsiACE The main branch now includes the new general error type. So you need to merge the main branch to resolve logic conflicts.

[leerho]

Waiting for @PsiACE

[PsiACE]

I have now completed the updates. This PR should be safe to merge. cc @tisonkun @leerho

[tisonkun]

LGTM. You may leverage the new codec utils in a follow-up PR.

[tisonkun]

nit: googletest has a contains matcher that can provide a better error message without manual construct.

https://docs.rs/googletest/latest/googletest/matchers/fn.contains.html

[tisonkun]

I think the cpp impl uses u64 for reverse_purge_hash_map's value. But let's review this point as a follow-up.

Read https://github.com/apache/datasketches-cpp/blob/7bb979d3ef8929e235bcd22d67579e1f695f6ecd/fi/include/reverse_purge_hash_map_impl.hpp#L346-L364 especially for how subtract_and_keep_positive_only replaces:

        self.adjust_all_values_by(-median);
        self.keep_only_positive_counts();

This is the only usage where i64 is needed in the current impl.

[tisonkun]

Cross-reference to #35.

I don't know whether this sketch works well in approx_topk scenario. I'll try.