Currently, passes are registered on a per-dialect basis, which
provides the smallest footprint obviously. But for prototyping
and experimentation, a convenience "all passes" module is provided,
which registers all known MLIR passes in one run.
Usage in Python:
import mlir.all_passes_registration