xr.align instead of xr.broadcast

[aaronspring]

there are two issues on xarray about xr.corr, from which we can improve xr.pearson_r and in general:

pydata/xarray#4804: use a, b = xr.align(a, b, exclude=dim) as broadcast has join="outer" which fills it with NA which then get ignored; align uses join="inner". I remember I implemented that, but I am no broadcasting/align expert. I was just happy tests passed...
Here we are also not sure why xs.pearson_r is faster than xr.corr

pydata/xarray#4768: I raised whether to implement weighted in xr.corr

[aaronspring]

also fix that probabilistic members are broadcasted (except member_dim)