Improve AI draw-punisher avoidance heuristics

[Madwand99]

This narrows the previous AI decision work (#10559) to a lightweight heuristic for avoiding dangerous draw effects into opposing draw-punishers.

Changes include:

• Adds a static heuristic for detecting risky self-draw plays without running a generic safety simulation.
• Detects draw-heavy effects such as wheel-style hand redraws and permanent-triggered draw engines.
• Detects opposing draw-punisher triggers by behavior, not card name, including damage/life-loss/token payoff triggers.
• Applies the heuristic to normal AI spell selection, simulation picker candidates, and optional free-cast decisions.
• Preserves a commander exception so the AI does not refuse to cast its own commander from the command zone.
• Fixes Timetwister AI logic matching to handle the card script’s casing.
• Adds regression coverage for Wheel of Fortune, Timetwister, Echo of Eons, Mindmoil, Teferi’s Puzzle Box, Arjun, Xyris, Nekusar, and Kederekt Parasite cases.

[tool4ever]

hmn, I'll have to think on this:
we usually don't attempt guessing this sort of two level deep
(e.g. we have helpers for direct damage from etb/casting but this is more "I might draw from it" + "then I might lose life from drawing")

I'm not convinced that while at first this might look good for a few specific cases it doesn't end up causing just as much problems at other times, especially when it's hooked up this high in the chain 🤔

[Madwand99]

Makes sense. This heuristic is intentionally trying to catch cases that are one step beyond direct “this spell hurts me”: “this play causes me to draw a lot” plus “an opponent has an active draw punisher.” That is why it catches things like Wheel effects into Nekusar/Xyris, but also why it is riskier than the usual direct ETB/cast damage checks.

I think there are a few possible directions:

1. Keep the current high-level hook. This catches the broadest set of cases, including future draw-engine permanents like Mindmoil, Teferi’s Puzzle Box, and Arjun. The downside is exactly what you pointed out: false positives are expensive because this blocks the play before normal AI evaluation can consider context.
2. Move the heuristic into narrower ability logic, e.g. ChangeZoneAllAi / wheel-style effects only. This is much safer and easier to reason about, but it would no longer catch permanents like Mindmoil/Puzzle Box/Arjun being cast into Xyris/Nekusar/Kederekt.
3. Split the heuristic into narrower cases: immediate wheel-style draw effects in the relevant ability AI, plus a separate permanent-spell check only when the card being cast has its own large self-draw trigger. That would still catch Mindmoil/Puzzle Box/Arjun, but avoid using this as a broad “before any spell” gate.

Happy to go with any of these options.