Cameraz

[AnastaZIuk]

Introduces the new camera stack as Nabla::ext::Cameras.

[Erfan-Ahmadi]

• this function is unused and is a bit funny, cause it's the same function but when you pass hlsl::matrix<T, 4u, 4u> it treats as row-major but otherwise column major and the function is never used btw.

[Erfan-Ahmadi]

the convention in your camera ext is super unclear:

• hlsl::matrix in nabla is always treated as row major (that's why we privately inherit from glm's)
• means mat3x3 { right, up, forward } will put those vectors in rows not columns
• you're using camera_vector_t and camera_matrix_t and so on, but it's just alias of nbl::hlsl version and add no value and only confusion
• you can see you were confused here in this comment and caused a bug because you assumed column-major matrix
• sometimes you're using mat3x3 and treating it as a column vector for storage only but sometimes you're passing those columns separately because you know it'll bite you in the ass if you used a matrix
• sometimes you're putting position in the last column correctly which works with mul. but sometimes you're treating last row as translation which is definitely incorrect, it mostly works because nothing ever muls with them. but the conventions are unclear and causes people to make bugs.

[Erfan-Ahmadi]

we have nbl::hlsl::RelativeApproxCompareHelper that should even work for matrices. this is redundant

[Erfan-Ahmadi]

not included in this PR but needs to be fixed:

https://github.com/Devsh-Graphics-Programming/Nabla/blame/89cfa0b58c4d719c3dd3a245f6b46f73c6caa638/include/nbl/builtin/hlsl/testing/relative_approx_compare.hlsl#L25

compare with 0.0 or any other subnormal even with a generous tolerance with return false. that's causing some quaternion stuff to be nan when they shouldn't

e.g here:

Nabla/include/nbl/builtin/hlsl/matrix_utils/matrix_runtime_traits.hlsl

Line 36 in 89cfa0b

orthogonal = orthogonal && testing::relativeApproxCompare(hlsl::dot(m[i], m[(i+1)%N]), scalar_t(0.0), 1e-4);