collate

• Should we update collate_batch so that it pads in both direction to the maximum shape ? So that if the number of individuals (rows) varies as well it works. It will be more generic (the current one pads to the max number of snps in the batch)

• Should the default be collate_batch with padding to batch max ? That's the current behavior. If we change to no padding there will be an error when we try to create batches of unequal sizes. @jcury says it's better so that people know what they do. I have no strong opinion. I find it convenient as it now though, it could raise a warning at the very beginning ?

• Add the collate function that pads to a given (n_ind, n_snp, pad_value) (should be an existing torch function) so that all batches have exactly the same size.

• How will people specify the collate function in the config file ?