Fix crash in armor coverage test

[BrettDong]

Summary

Bugfixes "Fix crash in armor coverage test"

Purpose of change

• Fix Crash in MinGW cross-compile CI test #54949

Tests have been consistently crashing on MinGW recently, and also crashes randomly in GCC 9 Linux configuration.

Describe the solution

The test case spawns two test NPCs, one of them overlaps with avatar character, causing very weird behaviors in projectile attack and damage handling code. This pull request shifts the position of test creatures by 4 tiles and seemingly solves the crash.

Describe alternatives you've considered

Testing

Wait for general build matrix to pass on all build configurations.

Additional context

[BrettDong]

This pull request also vastly shortens the time spent on armor coverage tests:

(~[slow] ~[.])=> 2.477 s: Full melee and ranged coverage vs. ranged attack
(~[slow] ~[.])=> 2.477 s: Ranged coverage vs. bullet
(~[slow] ~[.])=> 2.301 s: No ranged coverage vs. ranged attack
(~[slow] ~[.])=> 2.301 s: Ranged coverage vs. bullet

---

master branch:

(~[slow] ~[.])=> 104.525 s: Full melee and ranged coverage vs. ranged attack
(~[slow] ~[.])=> 104.525 s: Ranged coverage vs. bullet
(~[slow] ~[.])=> 105.030 s: No ranged coverage vs. ranged attack
(~[slow] ~[.])=> 105.031 s: Ranged coverage vs. bullet

[akrieger]

I'm kinda shocked by the test runtime reduction. Was it hitting some edge case in some pathing/visibility/aim algorithm or something?

[BrettDong]

I'm kinda shocked by the test runtime reduction. Was it hitting some edge case in some pathing/visibility/aim algorithm or something?

When I was stepping in the debugger to troubleshoot the use-after-free error, I incidentally found that the majority of time is spent on handling projectile attack damage on our avatar character rather than the test dummy NPCs.