Enforce start codon + fix bug explicitly setting training data on command line

[LanderDC]

Hi,

This PR handles #64 + I found a bug when specifying a custom training file for ORF prediction on a new sequence: even if a training file is specified, pyrodigal would retrain anyway and overwrite the training info.

pyrodigal/src/pyrodigal/cli.py

Lines 271 to 278 in afebe63

	if args.p == "single":
	# use the same interleaving logic as Prodigal
	sequences = list(parse(input_file))
	training_info = gene_finder.train(
	*(seq.seq for seq in sequences),
	force_nonsd=args.n,
	translation_table=args.g
	)

Fixed with:

if args.p == "single" and training_info is None:
...

What has changed:

• The -c argument now takes one of the options "none" (default), "start" or "both". For backwards compatibility, specifying only -c is equal to -c both which results in both a start and stop codon must be present.
• The GeneFinder class closed argument should now be a list of two boolean values (again for backwards compatibility, if only one is given the new closed_start and closed_stop arguments will take that value).
• Updated docs
• Added test for the enforced start codon + changed test_extract_edge_start() because the extract function does not take the closed argument anymore