`.rolling_exp` arguments could be clearer

Is your feature request related to a problem?

Currently we call .rolling_exp like:

da.rolling_exp(date=20).mean()

20 refers to a "standard" window type — broadly "the same average distance as a simple rolling window. That works well, and matches the .rolling(date=20).mean() format.

But we also have different window types, and this makes it a bit incongruent:

da.rolling_exp(date=0.5, window_type="alpha").mean()

...since the window_type is completely changing the meaning of the value we pass to the dimension argument. A bit like someone asking "how many apples would you like to buy", and replying "5", and then separately saying "when I said 5, I meant 5 tonnes".

Describe the solution you'd like

One option would be:

.rolling_exp(dptr={"alpha": 0.5})

We pass a dict if we want a non-standard window type — so the value is attached to its type.

We could still have the original form for da.rolling_exp(date=20).mean().

Describe alternatives you've considered

No response

Additional context

(I realize I wrote this originally, all criticism directed at me! This is based on feedback from a colleague, which on reflection I agree with.)

Unless anyone disagrees, I'll try and do this soon-ish™

Start at the rolling_exp entry point and trace how the date argument and window_type are interpreted. Review the existing rolling_exp tests, then define and verify the proposed dictionary form for non-standard window types while preserving the existing standard form.