Option to disable coercing in Model.eval?

[kmnhan]

Hello,

Although array coercing has a much more consistent behavior with the 1.2.2 update, it happens to break some of my existing code that relies on xarray's broadcasting.

For instance, when I have a ModelResult from a fit, I used to do something like result.eval(x=arr) where arr is a xarray.DataArray.
This would return a xarray.DataArray which retains the coordinates and attributes of arr. Now, the output is coerced into a numpy array.

Hence it would be great if the coercion for eval could be toggled off, for instance with a keyword argument, where result.eval(x=arr, coerce_farray=False) would have the old behavior. This would also mean more consistency with Model.fit.

[newville]

@kmnhan Can you just call your model function directly?

I suppose ModelResult.eval() could have the option to suppress coerce_farray. I'm not sure we want to do that.

I am somewhat sympathetic, and we do try to allow "array-like" data on input, but lmfit.ModelResult.eval is a method of the lmfit classes, and we need to get the output of eval() to be a real ndarray. But, we need solutions for #875 and #900.

So, I guess I am a "maybe" on this...