refactor(math): Implement Matrix4x4::Inverse, Matrix3D::Get_Inverse and replace unsafe Matrix4x4 to D3DMATRIX casts with conversion functions that apply the required transpose

[xezon]

• Fixes Properly implement transpose between Matrix4 and D3DMatrix #552
• Closes refactor(dx8): Replace unsafe Matrix4x4/D3DMATRIX casts with proper transpose conversion #1852

This change implements Matrix4x4::Inverse and Matrix3D::Get_inverse so that we do not need to use D3DXMatrixInverse for Matrix4x4 and Matrix3D anymore (there are 4 callers, verified to produce same result). Chat GPT was used to help implement the text book math.

Eliminates all unsafe C casts (that I was able to find) between D3DMATRIX, D3DXMATRIX and Matrix4x4, Matrix3D.

The following new helpers functions help converting between the matrices:

void To_D3DMATRIX(_D3DMATRIX& dxm, const Matrix3D& m);
_D3DMATRIX To_D3DMATRIX(const Matrix3D& m);
D3DXMATRIX To_D3DXMATRIX(const Matrix3D& m);

void To_D3DMATRIX(_D3DMATRIX& dxm, const Matrix4x4& m);
_D3DMATRIX To_D3DMATRIX(const Matrix4x4& m);
D3DXMATRIX To_D3DXMATRIX(const Matrix4x4& m);

void To_Matrix4x4(Matrix4x4& m, const _D3DMATRIX& dxm);
Matrix4x4 To_Matrix4x4(const _D3DMATRIX& dxm);

TODO

• Replicate in Generals

[xezon]

Generals will fail to compile until replicated.

[stephanmeesters]

What's the benefit of having two Matrix3D outputs here (no longer using void)? The only difference I see is that if m is singular then out is unchanged while the function outputs NULL, otherwise the function outputs a pointer to out