Cubic numba proper compilation

[zeffii]

solves the signature issue, does not solve auto compilation for later reuse.

• rewrites minor parts of the cubic function to match the subset of numpy that numba can reliably read
• adds decorator and faux decorator
• adds automatic signature generation (compilation step for diskstorage does not work yet!)

turns out that you can generate a signature automatically ( why the docs aren't explicit about this is baffling) by doing

import numba

def make_signature_tuple(*params):
    return tuple([numba.typeof(p) for p in params])

param_1 = np.array([[0.0, 0.1, 2.0], [1.0, 0.3, 2.0], [2.0, 0.1, 2.3]])   
param_2 = 3
param_n = np.array([0, 2, 3, 4, 12, 14])
params = param_1, param_2, param_n

print(make_signature_tuple(*params)
>>> (array(float64, 2d, C), int64, array(int32, 1d, C))

[zeffii]

with the caching argument set to True the function is definitely stored to disk (pickled..)

https://numba.readthedocs.io/en/stable/reference/envvars.html#numba-envvars-caching

using the sv wrapper for njit to store the result of njit locally in a function dictionary is faster (permanently!) than, purely importing numba.njit(cache=True) and applyin that to the function.

[zeffii]

@portnov what do you think of this PR. any improvement ideas?

[zeffii]

if we can get some consensus on this, i'm interested in applying numba to other areas too.

[zeffii]

I think this is the least invasive approach. njit is attempted if numba is found, else the function just passes through the wrapper unchanged.