Warning for DTI normalization

[rutgerfick]

The static 'min_signal' option set to 1 in the DTI fitting causes an error for normalized data.
As in normalized data there is only data between 0 and 1, the min_signal threshold sets everything to 1, giving a constant signal.

It would be a good idea to put a warning here that checks if the data is normalized. Otherwise people might be confused as to why all their eigenvalues are constant, while it is actually a setting in the fitting process that causes this.

[arokem]

Thanks for the suggestion - I've run into this problem myself too in some simulations I was doing. I took a look around the code and I don't think that it would be too hard to implement. You'll need to set a default min_signal value upfront for the TensorModel class, and then check the data when it enters the fit method.

Another option is to change the min_signal to something much smaller than 1. Say 0.0001. Might still work OK, no?

[arokem]

closed through #447