Parallelize some recursive options

Especially `shed_lint --urls` could seriously benefit from a task queue and a number of worker threads popping off that queue.

I imagine we could apply that same theory to a number of other similarly recursive operations where we're doing one "thing" per discovered repository.