Make behavior of `sample_many_runs` and `sample_cle` consistent

[manuelosos]

Currently, the arguments of sample_many_runs and sample_cle are inconsistent.

• sample_cle only takes a singular intital state as argument while sample_many_runs can take multiple.
• sample_cle uses by default multiprocessing while sample_many_runs does not.
• Arguments are named differently max_time in cle and t_max in sample_many_runs.
• No option to provide a seed in sample_cle
• sample_cle numba compilation is not cached.

[lueckem]

No option to provide a seed in sample_cle

As per the docs, the Generator object is not thread safe, so it will not be possible to pass one. It is probably possible to set an integer seed like this, but we should verify that it works with parallelization.

sample_cle uses by default multiprocessing while sample_many_runs does not.

We could make it single threaded by default, and it is even possible to set the number of threads in numba, see here.
But when would we ever want to not use parallelization in sample_cle?
For sample_many_runs it does make sense to limit the number of processes, since each process has a copy of the network in memory.