Not preloading covariance matrices causes error in crossval

[OleBialas]

When setting preload=False in the TRF instance, crossval produces an error. I don't have time to fix this right now, so I'll just leave a minimal example here:

from mtrf import TRF, load_sample_data
from mtrf.stats import crossval

tmin, tmax = 0, 0.4
regularization = 1000
stimulus, response, fs = load_sample_data(n_segments=10)

# This workds
trf = TRF()
crossval(trf, stimulus, response, fs, tmin, tmax, regularization)

# This does not
trf = TRF(preload=False)
crossval(trf, stimulus, response, fs, tmin, tmax, regularization)