Parallelise random walk

[ignasiialemany]

The code is optimised to run in serial, but if a large number of walkers and/or time steps is used the runtime can be a limiting factor. Splitting walkers between cores/threads allows for faster overall execution but requires refactoring of the code.

• Implement parallel random number generation using numpy to ensure reproducibility irrespective of parallelisation
• API changes to the run and one_step methods only via optional arguments (kwargs) that default to serial execution

[janniklasrose]

Added support for random number streams in #4. I suggest adding a ParallelMonteCarlo class that wraps around/extends/replaces the existing MonteCarlo class with the methods overridden to handle parallelisation

[janniklasrose]

The only caveat would be that results are not reproducible between runs with different numbers of threads, but I don't see a way around that without major refactoring anyway.