various updates to sweep pipeline

[mufernando]

Various edits related to the production run of the sweep pipeline.

• Changes to the SLURM oregon profile to make Snakemake play nice with the huge load of sims
• Increase in number of replicates per param (from 200 sims to 1000)
• Update in diploSHIC (new Keras version)
• Fixed a bug in diploSHIC training
• Changed training params in diploSHIC so that the training is not too far from the simulated data (with stdpopsim) -- this can be diagnosed with the pca_sims_training.ipynb
• Changes to resource usage in some rules
• write_vcf part of the pipeline had to be adapted to deal with infinite sites assumption (bumping variants at 0 to position 1; removing variants that may happen at last position after rounding).

[andrewkern]

is this ready to merge?