Remove underspecified vector_t

[gmlueck]

This is change 9 of 9 that fix problems with the specification of the vec class. An implementation that follows the existing specification would not accept common code patterns and would not pass the CTS. None of the existing implementations actually follow the existing specification.

This change removes the vector_t type alias from the vec class and also removes the conversions to and from this type. The intention of this type was to provide a conversion between vec and a compiler-specific builtin vector type. However, these conversion also lead to ambiguity errors in common code patterns such as:

sycl::vec<sycl::half, 1> h1;
sycl::half h;
h = h1;

Prior to this change, the last line could result in an ambiguity error from the compiler between:

• Convert vec<half,1> -> half, call half::operator=(const half&)
• Convert vec<half,1> -> _Float16, convert _Float16 -> float, call half::operator=(float)

What's worse is that the exact type alias of vector_t is implementation-defined, so these ambiguity errors could be different for each SYCL implementation.

During the WG meeting, we learned that the two major SYCL implementations (DPC++ and AdaptiveCpp) define vector_t differently even though they are both based on clang. Given this lack of commonality, the WG decided to remove vector_t entirely from the spec. Implementations can decide for themselves whether to provide a compiler-specific conversion like this.

These changes correspond to slides 29 - 31 of the presentation that was discussed in the WG meetings.