Detect attack-punisher enchantments by trigger instead of by name

[omraj21]

AiAttackController only recognised Revenge of Ravens, by name, when working out whether attacking would cost the AI life:

// TODO: detect Revenge of Ravens by the trigger instead of by name
revengeOfRavens = !CardLists.filter(player.getCardsIn(ZoneType.Battlefield),
        CardPredicates.nameEquals("Revenge of Ravens")).isEmpty();

Several cards are printed with the same trigger, and the AI walked straight into all of them:

card	trigger
Hissing Miasma	Whenever a creature attacks you, its controller loses 1 life
Blood Reckoning	same, for creatures you don't control
Marchesa's Decree	same, alongside the monarch ability

This does what the TODO asks and looks for the trigger: an Attacks trigger on the defending player's battlefield whose effect is LoseLife aimed at TriggeredAttackerController.

With the old name check, the added test shows the AI sending all three attackers into lethal life loss:

holdsBackAgainstHissingMiasma  AI should not attack into lethal life loss expected:<0> but was:<3>
holdsBackAgainstBloodReckoning AI should not attack into lethal life loss expected:<0> but was:<3>

What deliberately does not count

Cards that punish the attacking creature rather than its controller — Circle of Flame dealing 1 damage to the attacker, Raking Canopy, Palazzo Archers — are excluded. The creature pays that cost, not the player, and normal combat evaluation already accounts for it. There's a test asserting the AI still attacks into Circle of Flame, so the check can't quietly widen to cards it shouldn't cover.

Two smaller corrections in the same logic

• The life total is compared against the actual life lost per attacker rather than assuming 1, so a card draining 2 each is handled correctly.
• The "not worth attacking" filter uses that same number instead of a hardcoded getNetCombatDamage() <= 1.

Amounts that aren't plain numbers are skipped rather than guessed at, since overestimating here would make the AI refuse to attack for no reason.

Verification

5 tests: no-punisher control, Revenge of Ravens (the case the name check already covered), Hissing Miasma, Blood Reckoning, and the Circle of Flame exclusion. Two of them fail on the old code.

Full mvn -U -B clean test: 12 modules, 308 tests, 0 failures, run three times.

---

Written with the help of GitHub Copilot CLI; the commit carries a Co-authored-by trailer for it.