Do EWMean and EWVar make sense? Should a choice more consistent with Var and Mean be offered?

[microprediction]

I've found myself using fewvar.py in place of ewvar.py because the answers make more sense (to me) during the burn-in period and it converges to var.py as we vary the fading_factor. In contrast ewvar.py is highly sensitive to the first value and will take a long time to converge, potentially. Similar comments apply to the EWMean.

As an example, for some but not all uses the running mean of [1, 2] "ought" to be close to 1.5 in the limit of fading_factor -> 0. Whereas if we initialize at 1 and creep very slowly we'll be at 1.99 instead.

Maybe I'm missing something?

LMK if it make sense to add fewvar.py or fewmean.py to the package (ideally they would be in rust though).