Remove skip predict code in gaussian.Taylor as it might just work with `ignore_nan_dims=True`

[SamDuffield]

Probably we want to set ignore_nan_dims=True as default

See e.g. the code here

cuthbert/cuthbert/gaussian/taylor/non_associative_filter.py

Line 207 in b3b63cc

F, c, chol_Q = linearize_log_density(

[SamDuffield]

So this doesn't work

Running linearize_log_density with ignore_nan_dims=True will return F, C all nans and chol_Q with zeros everywhere but nans on the diagonal.
Running with ignore_nan_dims=True gives the same but chol_Q with nans everywhere.

In either case this will break when sent to e.g. cuthbertlib.kalman.filtering.predict which really needs for a no-op to see F=identity, c=0 and chol_Q=0.

I think it's fine to leave it as it is at the moment, but we'll keep the issue open if it we can think of a better way to handle skipped dimensions in predict.