Check cheap block requirement before expensive one

[MostCromulent]

CombatUtil.mustBlockAnAttacker builds the set of attackers a blocker is obliged to block. Two of its tests were in the wrong order, because they are not peers: one establishes whether an obligation exists at all, the other asks whether an existing obligation is excused.

attackerLureSatisfied answers the first. If the attacker has no unmet must-be-blocked effect then nothing obliges this blocker, and the attacker is dismissed. It answers from the attacker's own keywords, and it dismisses almost every attacker, because hardly any creature has such an effect.

getBlockCost answers the second. A blocking requirement never forces a player to pay a cost, so where blocking would cost something the requirement is excused. Answering it means walking every card in the static-ability source zones and rebuilding each one's static abilities, and it excuses almost nothing, because block taxes are rare.

The cost question only carries meaning once a requirement is known to exist, so asking it first computed an excuse for obligations that were not there — for every attacker, against every candidate blocker. Asking the cheap question first puts the two in the order the rules pose them, and since both are continue guards over side-effect-free reads, the attackers reaching the later checks are unchanged. The same shape as #11916.

The saving grows with the width of the combat. On sixty distinct creatures a side, the turn with the first attack takes 44800 ms instead of 80189 ms, and calls to CardState.getStaticAbilities over two turns fall from 1007936214 to 362382774. A turn with no combat is unchanged. On a seeded four-player game with stock commander precons the difference is inside run-to-run variance, so no claim is made there beyond the game logs being identical.

---

🤖 Generated with Claude Code

[tool4ever]

unfortunately this optimization may not survive in the future since Hanmac wants to refactor them into statics too anyway: #3307

but I guess we can still take it for now :/

[MostCromulent]

Further testing, this time with go-wide token decks — AI-vs-AI games between mirrored Goblin token decks, rather than the sixty-creature board above. Three repetitions per arm, arms alternated and the order reversed between reps:

workload	before	after	saving
token decks	41,930 ms	33,920 ms	8,010 ms	−19.1%
four-player commander	116,834 ms	113,812 ms	3,022 ms	−2.6%
ordinary constructed	3,763 ms	3,523 ms	240 ms	−6.4%

The percentages track game length as much as the size of the saving: the three ordinary games total under four seconds, against nearly two minutes for commander.

Measured with the change applied to current master, which includes #11946.