Bispectrum Multi-Index last axis

[SBouchard01]

As mentioned in #61, but kept for a later PR:

In the case of the Bispectrum, I'm pretty sure we can use xarray to make the last axis a multi-index element that contains k1, k2, k3 and would later allow us to apply filter on those

[SBouchard01]

@epaillas here is a simple implementation of this MultiIndex coordinate if you want to implement this on the bispectrum:

values = np.random.rand(4, 3000) # Main information to store, of shape (time, loc)

# Coordinates
times = pd.date_range('2023-01-01', periods=4)
x = np.arange(3000)
y = np.arange(3000)
z = np.arange(3000)

value = xr.DataArray(
    values,
    dims=('time', 'loc'),
    coords = {
        'time': times,
        'x': ("loc", x),
        'y': ("loc", y),
        'z': ("loc", z),
    },
)

But you can't apply filter on non-dimension coordinates (apparently)
I'm trying to figure out a solution

[SBouchard01]

Applying the filter on those coordinates is more complex than expected. I have asked for help on the xarray repo, I'm hoping they can get back to me soon (and hope they have a workaround we can use)

[SBouchard01]

An answer to a similar question, that we could use at the data-compression step would be to create a custom index for k1, k2, k3 - but the docs on this implementation is quite limited :/