Add in different ways of handling noise

[Bradley-Karat]

Need to change sigma calculation based on b-shell. Right now noise sigma is estimate as S0/SNR, but it should be Sshell/SNR. Sshell is given by the simulation in a particular microstructural environment
Currently the noise implementation is given by the user through the SNR inputs, which will then be added to the ground-truth signal as rician or gaussian noise. However, this can put more of a burden on the user as they have to find out what noise level is optimal to simulate with given their subject data, then input that as SNR into the toolbox.
Consider:

• GAN for learning the noise distribution from real data to then inject into noiseless simulated signal (i.e. the generator of the GAN learns the noise distribution)
• Rectified noise floor considering this is magnitude data (refer to SANDI toolbox for this)