Make summary grid have multiple levels

[Alphare]

The current summary grid can get pretty unwieldy when you have a lot of different benchmarks. This is a proposal for making it easier to read by nesting them in a tree-like fashion.

This assumes that you have a coherent hierarchy of benchmarks: a.b.c would be shadowed by a.b.

[Alphare]

I made it so the output was sorted alphabetically.

[pv]

I'd prefer to split to subgroups only if there are more than 5 (or something) benchmarks. The number of benchmarks per class apparently varies between projects, and e.g. the scipy benchmark grid gets very long with this change.

The CI test failures also appear real.