Primary driven index builds seem to have slightly different behavior than regular ones in that they appear to eagerly create a collection when building an index on a non-existent collection, even if the index can't be built, requiring this exclusion (patch failure example).

When an index build fails, the empty collection should not exist.

An interaction between index builds (currently only PDIB hits this behavior) and a replicated record ids (set_recordids_replicated.js) causes the above behavior, but it is a test only issue. This ticket now covers making set_recordids_replicated.js more transparent (i.e., not cleaning up if it shouldn't and cleaning up when it should, based on the type and outcome of the wrapped command).