Implement baseline combat, projectiles, spells, and monster AI

[zoeyrose]

Outcome

Provide a complete but deliberately narrow combat path that proves the common action, actor, item, scene, encounter, and protocol foundations.

Scope and invariants

• Implement basic melee, ranged projectile, and spell actions with target validation, range/line rules, protections, blocking, damage/healing, ownership attribution, and death processing.
• Represent projectile and delayed effects as deterministic simulation entities/events with captured identities and cancellation rules.
• Create bounded monster perception, target selection, path/action decision, leash/reset, and common action submission; AI cannot bypass player action rules.
• Define friendly-fire/PvP/owner/pet attribution and disclosure policy explicitly.
• Emit semantic telegraph/action/effect events for Rust presentation without server-owned animation or prose.

Acceptance criteria

• Scenarios cover hit/miss/block/protection, projectile travel, interrupted/dead targets, reflected/owned sources as required, monster reset, simultaneous actions, reward attribution, and reconnect.
• Player and monster AI use the same action service; architecture tests reject privileged damage mutation.
• Replay and seeded RNG reproduce outcomes exactly.
• One melee monster, projectile, and spell flow work in the M3 slice; advanced discipline issues layer on them.
• Load/tick benchmarks cover dense combat and bounded AI work.

Dependencies and parallelization

Depends on #17, #23, #25, #37, #41, #48, and #49. Navigation may begin with a bounded simple policy; advanced techniques remain M5.

Independent implementation

Implement from the preserved issue-level/product behavior and independently authored fixtures. Do not copy or mechanically translate GPL server/Python code or tests. Replacement code is MIT; consumed content and assets keep their own licenses.