Add missing x86 pcasts
Reference issue
What does this implement/fix?
Add some missing pcasts, namely:
- float->bool (SSE, AVX)
- int->double (SSE, AVX, AVX512)
- float->double (AVX512)
Also introduce simplified method for enabling pcasts with a default inheritable struct vectorized_type_casting_traits. Remove _MM_FROUND_NO_EXC bit from _mm256_cvtps_ph per !1362 (merged). Clean up array_cwise (including very annoying "condition always true" warning in MSVC).