origin wildcard does not support underscores

[huddlej]

Current Behavior
To define inputs for the workflow, we create a builds.yaml file that contains a list of named inputs in a format like this:

inputs:
  - name: input1
    metadata: data/example_metadata.tsv
    sequences: data/example_sequences.fasta.gz

The name field of each input is used in the Snakemake workflow in the origin wildcard. This wildcard includes a leading underscore (by design) such that the example input about would produce files like results/filtered_input1.fasta where the wildcard is _input1.

The constraints on the format of this "origin" wildcard do not allow for underscores in the input names. For example, the following reasonable input definition:

inputs:
  - name: getting_started
    metadata: data/example_metadata.tsv
    sequences: data/example_sequences.fasta.gz

produces the following unintelligible error:

MissingInputException in line 363 of /Users/jlhudd/projects/nextstrain/ncov/workflow/snakemake_rules/main_workflow.smk:
Missing input files for rule index_sequences:
results/filtered_getting_started.fasta

Expected behavior

Users should be able to define whatever names they like for their input data and have these names be processed by the workflow without any errors.

How to reproduce

Copy and paste the example inputs entry above into my_profiles/getting_started/builds.yaml.

Possible solution

Instead of placing the origin wildcard inside the name of each associated file, use the origin wildcard as a subdirectory in results/ (or even data/ might make more sense). Additionally, drop support for the deprecated config["sequences"] and config["metadata"] interface and require users to define at least one entry in the config's inputs.

These changes will allow us to know that we always have an "origin" defined (instead of supporting the optional empty origin) and they will also make the wildcard's values more flexible because they can contain any reasonable values that a directory name can have. This approach will also have the benefit of cleanly organizing files into subdirectories by input, allowing users to discover and inspect these files more easily.

cc: @jameshadfield for comments on this proposed solution. I'm happy to implement this.