Improve performance of lyap_e

[CSchoel]

So far, I have not received any complaints about that from users, but lyap_e is rather computationally expensive for larger datasets and there might be a few low-hanging fruits to improve it.

Most notably, there is no need to loop over every orbit vector in the outer loop. We just need to visit enough of them to achieve statistical stability. Maybe there should be a parameter to control that.

DoD

• [ ] Profile the performance of lyap_e.
• [ ] Test how both time and output accuracy behave when we cut the number of vectors to visit in half.
• [ ] If the test is positive, add a parameter to sample orbit vectors instead of visiting each one.