SCORU: improve refutation game PBTs

A follow-up issue for !5352 (merged)

The PBTs should include

• generating valid and invalid PVM proofs
• generating valid and invalid inbox proofs
• cases where both proofs are valid but they don't match one another
• cases where both proofs are valid but they don't match the refutation game states

and (at least) two new strategies to test different player behaviours

• Lazy, a player who is honest but commits prematurely (doesn't compute enough steps)
• Eager, a player who is honest up to the correct commit tick count, but adds extra dishonest steps after

Also tighten these tests up, properly add the inbox and sort out current flakiness (flakiness wasn't reproduceable). Investigate the slowness of the ArithPVM tests---are we doing some unnecessary computation?

The PBTs should make sure to include the case of a zero-tick commitment: see !5332 (merged)