feat: impl CpcSketch

[tisonkun]

This refers to #37.

I plan to implement the following steps:

1. (DONE) PairTable (for storing sparse data)
2. (DONE) CpcSketch without union
   1. Empty state with empty data
   2. Sparse state with pair table
   3. Hybrid state with dense vector
   4. Pinned state with dense vector (with ICON estimator)
   5. Sliding state with dense vector
3. (Done) Union
4. Serde

[tisonkun]

We may not implement the merge function as the Java/C++ impl for PairTable but find another way to do the two-way merge. This is because in Rust, it's impossible to hold a mutable ref when an immutable ref is still in used, which is how PairTable::merge is used in practice:

    PairTable.merge(srcPairArr, 0, srcNumPairs,
        allPairs, srcNumPairs, numPairsFromArray,
        allPairs, 0);  // note the overlapping subarray trick

The real effect here is to perform a two-way merge of allPairs[srcNumPairs..numPairsFromArray] and srcPairArr. There should be a more proper way to do this in Rust.

[tisonkun]

This PR is now ready for review.

It's mainly ported from the datasketches-cpp impl, so I tag @AlexanderSaydakov as a potential reviewer.

Union and serde (compression) would be implemented as follows. But the current state is a reviewable & mergeable minimal feature set.

[tisonkun]

To be removed when Union and Serde get implemented.

[tisonkun]

False positive when iterating over index can be more expressive.

[tisonkun]

Java uses std Arrays.sort here. We may use [T]::sort_stable (or unstable?) as well. But this is how C++ impl does.