[Refactor Suggestions I] Mixins all the way

[CodyCBakerPhD]

First of several draft proposals opened for comparison; each offers advantages and disadvantages in terms of readability and comprehension (dependent on knowledge of package structure)

This one pushes the mixins to the ultimate level, so that each actual benchmark run by ASV is only comprised of the specification of that particular instances attributes

The changelog view is a little muddled but the direct file view showcases how easy it is to read: https://github.com/NeurodataWithoutBorders/nwb_benchmarks/blob/1f58d3830e53f5177be97088355f05a59e1fdf63/src/nwb_benchmarks/benchmarks/streaming.py

However, this approach comes with a lot of implicit understanding of how the mixins are separated, with a lot of redirection required to understand what any single benchmark is actually doing whenever it runs

But it does make it really easy to expand a matrix of cases by taking cross-combinations of setups and time tests

Also beginning to adopt sklearn import structure since centralized imports are now being used

[CodyCBakerPhD]

The implicit knowledge of how to use this PR is the most of any of them, but the pattern can be learned and documented

Each benchmark is a mixin of the structure ChildBenchmark(BaseBenchmark, BaseSetup, ExtraMethodsUsedByEitherBase)

where BaseBenchmark defines all the tracking tests to run, and BaseSetup defines the setup and teardown methods to occur around those tracking tests; both of these might rely on interchangeable intermediate methods that are specified by the extra mixins

The resulting benchmark file is very easy to read, but takes a lot of redirection (one per mixin) to understand exactly what is is running

[CodyCBakerPhD]

Note that the main benchmarks folder is for things ASV will actually point to; base class definitions should probably be outside of this

Analogous to having a src/testing submodule that lives outside of an outer tests folder when using pytest