PhysicsDirectSpaceState2D/3D: reusable strongly typed results

[io-mi]

This PR refactors the query methods in PhysicsDirectSpaceState2D/3D to use strongly typed result objects instead of allocating result containers on every call. This aligns the API with body_test_motion() in PhysicsServer2D/3D and improves performance, though it is backward-incompatible.

API changes

Old	New
Dictionary intersect_ray(parameters: PhysicsRayQueryParameters2D)	bool intersect_ray(parameters: PhysicsRayQueryParameters2D, result: PhysicsRayIntersectionResult2D)
Array[Dictionary] intersect_point(parameters: PhysicsPointQueryParameters2D, max_results: int)	bool intersect_point(parameters: PhysicsPointQueryParameters2D, result: PhysicsPointIntersectionResult2D)
Array[Dictionary] intersect_shape(parameters: PhysicsShapeQueryParameters2D, max_results: int)	bool intersect_shape(parameters: PhysicsShapeQueryParameters2D, result: PhysicsShapeIntersectionResult2D)
PackedFloat32Array cast_motion(parameters: PhysicsShapeQueryParameters2D)	bool cast_motion(parameters: PhysicsShapeQueryParameters2D, result: PhysicsShapeCastResult2D)
Array[Vector3] collide_shape(parameters: PhysicsShapeQueryParameters2D, max_results: int)	bool collide_shape(parameters: PhysicsShapeQueryParameters2D, result: PhysicsShapeCollisionResult2D)
Dictionary get_rest_info(parameters: PhysicsShapeQueryParameters2D)	bool get_rest_info(parameters: PhysicsShapeQueryParameters2D, result: PhysicsShapeRestInfoResult2D)

Equivalent changes are applied to the 3D.

New classes

• PhysicsRayIntersectionResult2D/3D
• PhysicsPointIntersectionResult2D/3D
• PhysicsShapeIntersectionResult2D/3D
• PhysicsShapeCastResult2D/3D
• PhysicsShapeCollisionResult2D/3D
• PhysicsShapeRestInfoResult2D/3D

Motivation

1. Strongly typed access

The new typed result objects eliminate the need to manually declare types when accessing fields:

var result := PhysicsRayIntersectionResult3D.new()
if world.direct_space_state.intersect_ray(params, result):
    var normal := result.get_normal()
    var position := result.get_position()

Old usage required explicit type annotations:

var result := world.direct_space_state.intersect_ray(params)
if result:
    # Cannot infer the type of "normal"
    # var normal := result["normal"]
    var normal: Vector3 = result["normal"]

2. Performance

Result objects can be created once and reused:

var result := PhysicsShapeCastResult3D.new()
for i in range(1_000):
    if world.direct_space_state.cast_motion(params, result):
        ...

Avoids per-call result allocation and improves performance.

3. Consistency

This matches the PhysicsTestMotionResult2D/3D and PhysicsTestMotionParameters2D/3D pattern already used by body_test_motion().

Crude microbenchmark comparisons (3D)

• intersect_ray()

• intersect_point()

• intersect_shape()

• cast_motion()

• collide_shape()

• get_rest_info()

Notes

• Possibly related to #112433

[AThousandShips]

Thank you for your contribution!

This is a pretty majorly breaking change, is there a proposal to track this or to evaluate support?

This would benefit from having a proper proposal created here to discuss the details and evaluate support for this, as it would be something that would need good support to justify the changes and the need for them

I'd say backwards compatibility needs to be maintained by creating new methods instead of replacing the existing ones, as this would be very disruptive to anyone using the existing interfaces