Spend some time optimizing the inner heads() loop, approx 2x faster.
Various tweaks and their effect:

# Baseline
Found 25382 nodes, loaded in 2.194s
      7581 combinations
Known: 1.626s
Known (pyx): 0.993s
ratio: 1.6:1 faster

# Tweaks for NULL_REVISION and PyDict_Size, min_gdfo as long
Known (pyx): 0.945s
ratio: 1.7:1 faster

# PySet_Contains, PySet_Add
Known (pyx): 0.846s
ratio: 1.9:1 faster

# Avoid pop & push
Known (pyx): 0.564s
ratio: 2.9:1 faster

# pending_pop
Known (pyx): 0.542s
ratio: 3.1:1 faster