Fix incorrect return type cast

[audeck]

Currently, _set_limit_clamp_position() returns a Vector2i when a limit target is applied, resulting in target_position always being truncated to an integer. This, without any follow_damping, results in the truncated target_position propagating to the camera's global_position on line 577, possibly causing misalignment between the camera and the node it's following, which can be at float x and y values.

This pr simply changes the return type to Vector2, leaving any logic intact.

[ramokz]

As per #268, looks good to me!

[ramokz]

Thanks for spotting and updating this ✨