Improve Training data simulation

[gwarmstrong]

Some ideas of things that can/should be done

• ensure that reads are sample from the same location only once Ensure that reads are sampled from the same location only once in simulation #27
• Add simulated errors to training data (could involve neural network) train on data with errors #1
• Add error correction module to fix low-quality sequences (could use some real data with alignments to genome?) Add error correction module to fix low-quality (or ambiguous) positions in reads #28