BENCH: collect low-dependency asvs

[jbrockmendel]

The idea here is to collect benchmarks in such a way that we can say "This PR doesn't change code in files [...] -> We can skip running asvs in files [...]". (putting together a POC script to automate a bit of this)

There is a cost to refactoring benchmarks, since we have to re-run the new benchmarks on older commits (and im not sure if that happens automatically or if some human intervention is needed cc @TomAugspurger ?). On the flip side, this is the kind of problem that we can throw hardware at, so I don't worry about it that much. Still, the bar for merging is higher than it would be for non-benchmark refactors.

With that in mind, the other refactors I'm looking at:

1. We have benchmarks for e.g. isin scattered around, benchmarking [algos.isin(x, y), pd.array(x).isin(y), Index(s).isin(y), Series(x).isin(y)]. I'd like to collect these in one place and parametrize them.

2. Similar for Index methods, we have a lot of similar benchmarks for different index subclasses that id like to collect+parametrize.

[jreback]

seems fine can you merge master

[jbrockmendel]

rebased + green