Other norms for fitting? (feature request)

[willsharpless]

Hi thank you for making this awesome package,

I understand that several works have noted that the L2 norm may not be the best choice for fitting certain functional spaces. Personally, I have an application where the maximum error (Linf) is more important than the mean error (L2). One may also want a sparse fit in some cases too or a lasso regression. Would you consider adding minimization (via linear programming) over other norms as a possibility?

I see this package currently is using pinv and avoiding optimization, but perhaps code from pysindy might make this feasible?