feat(expr): Make intersects() return the intersection when non-empty

[effigies]

This updates the intersects() function in the schema expression language to:

1. Return the intersection, if non-empty. The order will match the order in the longer list.
2. Return null on empty intersections to ensure that intersects() remains falsey.

This seems like the simplest resolution for resolving bids-standard/bids-specification#914 along the lines of:

If participants.tsv exists, then the participant_label entries are a superset of subject directories and participant labels found in phenotype/.

With this change, the rule could be written:

SubDirSubset:
  selectors:
    - path == '/participants.tsv'
  checks:
    - |
      allequal(
        sorted(intersects(dataset.subjects.participant_id, dataset.subjects.sub_dirs)),
        sorted(dataset.subjects.sub_dirs)
      )

PhenotypeSubset:
  selectors:
    - path == '/participants.tsv'
    - dataset.subjects.phenotype
  checks:
    - |
      allequal(
        sorted(intersects(dataset.subjects.participant_id, dataset.subjects.phenotype)),
        sorted(dataset.subjects.phenotype)
      )

If we require that validators sort the arrays in dataset.subjects on load, then we can drop the extra sorted() calls here.

Because we've seen datasets with subjects in the thousands, I've updated this function to create a Set() based on the smaller list. There is very likely a small n where this constitutes a performance regression.

[effigies]

cc @ericearl

[effigies]

xref bids-standard/bids-specification#2044