reduce wnn runtime by fetching the precomputed no_batch_*

[bio-la]

in the current version of the integration , if wnn is run on no-batch corrected modalities, it will run neighbours on each modality on the flight in a "no_batch" way (i.e. on precomputed dimred such as PCA or LSI if specified) with the same param as specified for each of the no_batch unimodal analyses.
it's a different behaviour when wnn is calc on pre-batch corrected unimodal data, cause in that case the pipeline expects each batch corrected object to exist and it's correctly reflected in the decorators flow.

we need to modify wnn to fetch precomputed no_batch instead of running on the flight to reduce the runtime (currently runs nobatch twice per modality if wnn is called on no_batch)