Speed-up Genomic Benchmarks dataloader

[cbirchsy]

I've modified the dataset class such that sequences are simply stored in a list and indexed from memory rather than being read from file. With this change I observed a speedup in epoch time of up to ~ 5x

Dataset	Epoch time (old)	Epoch time (new)
dummy_mouse_enhancers_ensembl	3s	2s
human_enhancers_cohn	42s	14s
human_nontata_promoters	49s	10s
demo_coding_vs_intergenomic_seqs	2m 6s	25s
demo_human_or_worm	2m 19s	25s
human_enhancers_ensembl	3m 56s	1m 20s
human_ocr_ensembl	4m 48s	1m 32s
human_ensembl_regulatory	8m 21s	2m 37s

[exnx]

this looks great, thanks so much!