Implement Native Lua Vector Datatype

[MohabCodeX]

Replaces #1728
Fixes #321

This is basically the last part of the OOP/vector optimization work from #5278, #5279, #5280 and #5290. It completes the native vector idea that was originally started by @Pirulax in #1728.

The main goal here was to make vectors actual native Lua values instead of relying on userdata/metatables and C function calls. With these changes in place, OOP element properties are now faster than procedural calls (6ms vs. 41ms), while native vector math is 12.6× faster than legacy OOP vectors, getting close to raw CPU performance.

What changed

• Added vector / LUA_TVEC as a native 128-bit 4D floating point type in the Lua runtime.
• Arithmetic such as +, -, *, /, ^, unary - and length calculations now have VM fast paths, without going through metamethods or C functions.
• Supports .x, .y, .z, .w, numeric indexing ([1] to [4]) and swizzling like .xy, .xyz, etc.
• Native vectors are supported by CScriptArgReader, CLuaArgument event serialization and the JSON converters.
• Added the usual constructors and helper functions through the global vector library.

Lua API

Creating a vector:

vector vector ( float x, float y, float z [, float w = 0.0 ] )

x, y and z are required. w is optional and defaults to 0.0.

Example:

local pos = vector(0, 5, 3)
local target = pos + vector.forward * 10

print("Target position: " .. target)
print("Distance: " .. pos:distance(target))

Available methods

Both method and static versions are available:

• vec:dot(otherVec) / vector.dot(v1, v2)
• vec:cross(otherVec) / vector.cross(v1, v2)
• vec:length() / vector.length(v)
• vec:lengthSquared() / vector.lengthSquared(v)
• vec:normalize() / vector.normalize(v)
• vec:distance(otherVec) / vector.distance(v1, v2)
• vec:distanceSquared(otherVec) / vector.distanceSquared(v1, v2)

Constants:

• vector.zero = vector(0, 0, 0, 0)
• vector.one = vector(1, 1, 1, 1)
• vector.forward = vector(0, 1, 0, 0)
• vector.right = vector(1, 0, 0, 0)
• vector.up = vector(0, 0, 1, 0)

Benchmarks

500,000 operations per test:

Test	Inline OOP	Native SIMD vector	Improvement
Vector arithmetic (+, -, *)	2,918 ms	230 ms	12.7x
Vector math (:dot, :length)	263.6 ms	56.3 ms	4.7x
Component reads (.x, .y, .z)	695.3 ms	21.3 ms	33x
Total	3,877 ms	307.3 ms	12.6x

Sounds a bit unrealistic? Yeah, test it yourself :)))
vector_benchmark.zip

For comparison, running around 1.5 million synchronous operations in a single frame previously froze the game for roughly 4 seconds and triggered the black loading screen. With native vectors, the same workload causes only a frame drop of around a second or less, so normal gameplay shouldn't be affected.