ENH: add "adjust" parameter to ewmvar(), ewmstd(), ewmvol(), ewmcov(), and ewmcorr()

[seth-p]

Currently ewma() has an adjust parameter (affecting how the weights are calculated) whereas all the other ewm*() functions do not. When they call ewma(), adjust is implicitly set to its default value, True. This is probably fine in most cases, but if ewma() is going to expose the adjust parameter, the other ewm*() should as well.

For clarity, here's the description of the adjust parameter in _ewm_notes:

When adjust is True (default), weighted averages are calculated using weights
    (1-alpha)**(n-1), (1-alpha)**(n-2), ..., 1-alpha, 1.

When adjust is False, weighted averages are calculated recursively as:
    weighted_average[0] = arg[0];
    weighted_average[i] = (1-alpha)*weighted_average[i-1] + alpha*arg[i].

On a related note, I'm not sure what the policy/practice is for reordering parameters, but if it's not too strict I'd like to reorder the ewm*() parameters just a little so that (i) freq and how are next to each other, and (ii) adjust and ignore_na are next to each other.

[seth-p]

Note that even though all the ewm*() functions other than ewma() lack an adjust parameter, their doc strings say that they have one. (This is from the definition of _ewm_kw.)

[seth-p]

CC'ing @snth and @jaimefrio, who appear to have worked on this/related code.

[seth-p]

Along the same lines, should the rolling_var/std/cov() and expanding_var/std/cov() have a ddof parameter the way Series.var/std() do (though Series.cov does not)? They're all labeled "unbiased", and appear to effectively use ddof=1, so nothing is wrong, but I don't see why the treatment of ddof can't be uniform.

And similar my comment above on parameter order, for these functions too I'd like to reorder things a bit, e.g. so that freq and how are next to each other...