Check numba speedup

[grzanka]

• add another implementation of solver, just for testing numba code. This would probably violate DRY as I expect to have a plain copy of the current python implementation. In other words as the first step make a 1:1 copy of the python implementation from Rewrite initial_recombination.py in pure python #38 and call it numba_something
• add another entry for numba in solver abstraction layer
• add a short test suite for numba (or include numba in tests mentioned here Rewrite initial_recombination.pyx to pure Python #28 (comment)
• check numba speedup after refactoring of python code (i.e. extracting inner loop code into separate method)
• check numba speedup for plain python method adding various types of decorators

For numba testing, I suggest moving to Ares PLGRID computing node with 48 cores.

[grzanka]

For testing strategy evaluate and write down in the issue comment here:

• @jit speedup, I suspect similar speedup as for Cython
• automatic parallelization with @jit https://numba.readthedocs.io/en/stable/user/parallel.html#numba-parallel
• fast math https://numba.readthedocs.io/en/stable/user/performance-tips.html#fastmath
• use scipy/numpy compiled against MKL and check numba speedup https://numba.readthedocs.io/en/stable/user/performance-tips.html#linear-algebra

[grzanka]

Consider removing if debug in numba version as numba may be much slower in the code containing conditional statements