Fix NEON bug in convert_vec3_to_vec4W0 and reenable tests

[alusch]

While testing #1353 on my Mac I incidentally got gtc_type_aligned.cpp to run 😅

First issue is a compile error in convert_splat:

error: implicit conversion increases floating-point precision: 'float' to 'double'

Switching to a plain 0 constant seems to make things happy on my machine, though if that causes problems elsewhere we could use {} T() instead.

Next was a test failure in test_copy_vec4_vec3 where xyz0(vec3(1, 2, 3)) was returning (0, 2, 3, 4) instead of (1, 2, 3, 0)! This turned out to be a simple flipped mask order in func_common_simd.inl.

With these changes the tests now run and pass on my M1 Mac, and hopefully also CI.

[ZXShady]

Be aware that {} casting is only usable in C++11