new regressions Huber and Theil-Sen for trendline

[martins0n]

Hi! I've added two robust regressions for trendlines. There's an import from scipy, maybe it's not great. But I think that robust regressions will be helpful for analyzing data.

[nicolaskruchten]

Thanks for this PR! We've been thinking about an extension mechanism for this "trendline" feature for a while, as an alternative to baking in an increasing number of types of regressions directly into the library. I was thinking of having this argument accept a function with a signature like (x_in, y_in) -> (y_out, results, hover_string) or something like that, so users like yourself would be able to define/tweak/pass-in different regression types.

The advantage to this function approach is that you could do something like the following:

def huber(t):
    def huber_impl(x_in, y_in):
        # math that depends on `t`
        return y_out, results, hover_string
    return huber_impl

px.scatter(..., trendline=huber(t=2.222))

In addition to this, I'd want to include some kind of px.trendline_functions dictionary where users could register "named" functions, and then refer to them by name like "huber" or something like that.

As our first new-regression-function contributor, how does this sound as an API?

[martins0n]

Use of decorators is a great and flexible option. But it's maybe not very obvious for users.
How to make it more obvious keeping it simple I have no ideas unfortunately. Maybe good examples would help.

[nicolaskruchten]

Thanks again for taking the time to submit this PR! I'm going to close it now in favour of #2418 but I will re-use this code when documenting the new pluggable interface, as examples of the kinds of things that can be done with it :)