implement pipeline dependency

[adelavega]

Certain pipelines, such as UMLS, operate on both the original text inputs as well as derivative outputs from participant_demographics

We need to implement a way to specific which inputs to run/execute on. This will be the responsibility of the meta-runnner script.

Pipelines that require other inputs will denote with something like:
REQUIRES = ['participant_demographics']
and have a similarly named argument

To determine which inputs to iterate over (i.e. which derivative versions), a few selection rules:

• If "latest", it will run latest output from that dataset slice + previous pipeline
• Otherwise, the configuration file can specify the arguments to select the inputs (i.e. which arguments to filter on for previous pipeline)
  • e.g. only select GPT-4 outputs, but get the latest one only