Improve AI multiplayer counterspell discipline

[Rauglothgor]

Summary

• avoid optional AI counterspells against targeted interaction between two other opponents in multiplayer games
• inspect stack target choices for the root ability and sub-abilities
• keep AI-threatening, self-buffing, mixed-target, non-targeted, and mandatory cases eligible

Notes

This is a narrow follow-up inspired by the multiplayer interaction concerns in #10514. It intentionally avoids broad board simulation or unrelated targeting rewrites.

Validation

• mvn -pl forge-ai -am -DskipTests compile

[tool4ever]

Are you an AI agent just randomly cherry picking? 🤔

[Rauglothgor]

Yes, Codex assisted with this PR.

This was not intended as random cherry-picking. I split out the CounterAi-only multiplayer interaction behavior from #10514 because #10514 touches several areas, including ChangeZoneAi, GameSnapshot, and tests. The goal here was to offer a smaller reviewable slice: optional AI counterspells should avoid targeted interaction between two other opponents, while preserving mandatory, self-threatening, mixed-target, and non-targeted cases.

If this slice is not useful separately, I'm fine closing it and leaving the broader direction to #10514. If a narrower or different cut would be easier to review, I can adjust it.

[tool4ever]

meh, it's not the worst idea
just might lead to merge conflicts or deviation from author intention if I have more comments here instead
though I don't really need to bother with explaining stuff too deeply for AI anyway 🤷‍♂️