Report effective tolerance for solutions

[maedoc]

inspired by #73 but realizing adaptive step sizes break a lot of assumptions for parallelizing solvers, it would be lower hanging fruit to incorporate some statistics on effective tolerance e.g.

• how many time steps exceeding error tolerance
• quantiles on tolerance for instance median, 90, 95% e.g. https://stats.stackexchange.com/a/445714

in most use cases, the user can then handle cases where tolerance was exceeded or quantiles are too high by retrying those simulations with lower dt.