[DOC] Simplify bench-ann scripts and documentation

[cjnolet]

The current bench-ann scripts and workflow is pretty low-level and not as simple to use as it could be. One of the standards for benchmarking ANN algorithms in Python is the popular ann-benchmarks library- the scripts are very easy to use, the workflow is very straightforward (specifying training/search params, querying/using available algorithms and datasets, plotting). I think our bench-ann has all the available features but they could presented in a manner which is more friendly to the end-user who just wants to compare the perf of our algorithms with others and doesn't necessarily want to jump through hoops to do so.

I propose we write scripts that mimic the ann-benchmarks as much as possible so that users already familiar with that repository can more easily use our stuff. This is somewhat how the big-ann-benchmarks scripts functioned as well.