• Move functions voxel_fit_function, curve_fit_wrapper, and fit_voxel from utils.py to lookclocker.py (because they are only used here).
• Move functions estimate_se_free_relaxation_time and T1_lookup_table from utils.py to mixed.py
• Refactored Mixed module to use classes similar to LookLocker (see Test looklocker #39)
• Implement test for concentration round trip, i.e start with a known concentration compute t1/ r1 values and then concentration again. To do this I also had to implement a few more functions e.g R1_from_concentration_expr and T1_from_concentration_expr.
• Add an example of how to propagate noise through the concentration process, e.g start with a known concentration -> compute the t1 value -> compute the looklocker / mixed signal -> add noise (assuming some signal to noise ratio(SNR)) -> compute new t1 value -> compute new concentration. This way the user can play with how the SNR affects the uncertainty in the concentration estimates.

I also improved the typing in some of these function. For example consider the following function

def concentration_from_T1_expr(t1: np.ndarray, t1_0: np.ndarray, r1: float) -> np.ndarray:
    return (1.0 / r1) * ((1.0 / t1) - (1.0 / t1_0))

According to the type-annotation, t1 needs to by a numpy array. However we know that t1 can also by a float, but in that case t1_0 and the return value should also by a float. We can specify this using a type variable as follows

import typing

T  = typing.TypeVar("T", np.ndarray, float)

def concentration_from_T1_expr(t1: T, t1_0: T, r1: float) -> T:
    return (1.0 / r1) * ((1.0 / t1) - (1.0 / t1_0))

Here we also explicity bound the type variable to either a numpy array or a float (meaning that these are the only accepted types).