auto-regression: xarray wrapper for predict

[mathause]

After #150 we should write a xarray wrapper for the numpy based auto regression prediction. It is not yet entirely clear to me how this should look, though.

[mathause]

_fit_auto_regression_xr and _draw_auto_regression_correlated_np don't work together super smoothly yet. This is how we can make it work:

import mesmer
import numpy as np
import xarray as xr

air = xr.tutorial.open_dataset("air_temperature")

# a 2D array is required - would be a flattened (unstructured) grid
# for simplicity use one row of longitudes 
data = air.air.isel(lat=0)

params = mesmer.stats.auto_regression._fit_auto_regression_xr(data, dim="time", lags=2)

mesmer.stats.auto_regression._draw_auto_regression_correlated_np(
    intercept=params.intercept.values,
    coeffs=params.coeffs.T.values, # unfortunate...
    covariance=np.diag(params.covariance), # diag to ensure cov is invertible 
    n_samples=3,
    n_ts=50,
    seed=0,
    buffer=20,
)

[mathause]

_fit_auto_regression_xr and _draw_auto_regression_correlated_np don't work together super smoothly yet.

This already works better after #306. Maybe we should also transform the coeffs?

What bugs me, though - how to we populate the resulting DataArray how can we pass dimension names and coordinates? This is further complicated by the fact that I expect coords for the time dim but not for the sample dim. Also is sample a good name?

Further (and related) ArmaProcess.generate_sample allows to generate nD arrays.