Imposing boundaries on b-values (maximum, minimum) should be done by dataset (not model)

[oesteban]

The DWI dataset class should be responsible for deciding what's a b=0 and what are valid b values (given the minimum threshold) and likewise, saturate (given a maximum b) or manipulate (e.g., cluster b-values of a DSI into shells, clean up effective gradients that are close to the nominal value of a shell but not exactly) b-values when necessary.

I'm removing this feature from model in #62.

[oesteban]

Similarly, adaptation of RAS+b and IJK+b should be handled elsewhere.

[jhlegarreta]

Maybe a Filter should be in charge of doing this? I took the latter approach in PR #71.