feat: add additional counting aggregations to AggBy

[lbooker42]

New aggregations are:

• AggCountNonNull()
• AggCountNull()
• AggCountNegative()
• AggCountPositive()
• AggCountZero()
• AggCountNaN()
• AggCountInfinite()
• AggCountFinite()

[devinrsmith]

I'll post the same sort of concerns here that I have for #6270; I think we would do better to express this in terms of a Filter. What if I want to count NULL or NAN or INFINITE? Or any combination?

[lbooker42]

Interesting idea, would prefer to see this as a new feature suggestion.

These agg_by ops are companions to Numeric / Basic vector ops (and will be paralleled by update_by cum_count and rolling_count variants) and this filter idea isn't trivial to extend to all these routines.

[lbooker42]

Note on extensive use of lambda and performance

I ran repeated tests on 4 different VM (arm64 on Mac M1) and verified that the lambda-heavy ops are not penalized and the lambda are (apparently) in-lined by the JIT.

The comparison is with AggSum that uses type-specialized operators but evaluates every value, AggCountNonNull that uses lambdas and evaluates every value, and AggCountAll (which aliases AggCount) that evaluates no values, simply returning the size of the bucket rowset.

The expected (and confirmed) result is that the new counting ops will fall between AggSum and AggCountAll/AggCount in performance.

[jmao-denver]

Python changes LGTM