Implement conditional requirement for required keys anywhere in the args spec

[phargogh]

When writing out the MODEL_SPEC for UNA, I inadvertently wrote {"required": <expression>} for vector and CSV columns, forgetting that validation of conditional requirement wasn’t yet implemented. Here’s what I’m thinking for a solution:

1. Create a copy of args where all the required keys (regardless of where it is in the nested dict) are checked for expressions. If an expression is found, evaluate it like we evaluate required keys at the top level. The result is bool values where the expressions were before.
2. This new dict could be validated as we validate it now since we already check for required: True when validating individual inputs.

Related to #722
Needed for #1150

Estimated time remaining

To be safe (and to make sure this is tested), I'd guess this is several days of effort, maybe 1 week of development time.

[phargogh]

Not quite anywhere in the args spec, but #1511 does implement conditional requirement for vector fields and for CSV columns and rows.