Feat/ai controller

[GianRathgeb]

When merged this pull request will:

• Introduces a modular AI system (AIController, AIAgent, ScenarioTaskRunner + scenes) that deterministically executes authored task chains.
• Ingests ScenarioData.tasks and builds per-unit queues by following prev_index/next_index, normalizing task shapes.
• Implements task semantics for Move, Defend, Patrol (cycle/ping-pong), Wait (with until_contact), SetBehaviour, and SetCombatMode.
• Maps behaviour to movement parameters and routes ROE to CombatAdapter with hold/return/open fire, including a short return-fire window.
• Wires AIAgent to existing sim via adapters/OrdersRouter so AI issues real game orders rather than moving nodes directly.
• Adds mission bootstrap to register enemy units, seed initial behaviour/ROE, and start AI control on mission load.
• Provides minimal LOS contact feed for until_contact waits and ROE “return fire” triggers.
• Includes optional debug hooks (signals/HUD overlay notes) and makes the codebase GDScript-4.5 linter-clean with strict typing.

Issue ticket number and link

Create AI Controller
#69

Checklist before requesting a review

• I have performed a self-review of my code.
• I have tested my code.
• I have documented my code in accordance with gdscript_documentation_comments.
• I have requested a review in discord.
• I have moved the related issues to the review section in the SCRUM Board.

[GianRathgeb]

The issues from the code review are now fixed, please review again

[GianRathgeb]

the issues with the AI should now be fixed. During the fixing of the issues, I created more issues which I then had to sovle. Here is everything I did:

• Added inspector-assigned AIController node in HQ table scene and injected adapters through exported refs so agents always find Movement/Combat/LOS/Router nodes.
• Simplified AIAgent adapter binding and behavior/ROE initialization to prevent null movement intents.
  Normalized scenario task queues and refreshed unit-id caches before AI registration to keep return-fire logic consistent across machines.
• Emitted engagement_reported(attacker, defender, damage) only when real damage occurs and updated all listeners to accept the new signature, stopping premature reveals.
• Ensured calculate_damage results are treated as Variants to avoid float .get() errors.
• Scaled LOS detection by terrain spotting multiplier plus target behavior modifiers so units can’t be spotted from unrealistic ranges.

[GianRathgeb]

• Added async triggers for AI units
• Added default behavior for AI units if no task is defined