Unify kernel dispatch routine around `halide_type_t` args[]

[slomp]

There are two ways in which halide runtimes implement their "run" routines. In one of them, an array of arg sizes is passed, while for the other, an array of halide_type_t is passed. Why not just unify it around halide_type_t since it is more expressive?

metal: #7505

cuda:
opencl:

[steven-johnson]

Agreed that the latter is better. Where are we using just sizes?

[slomp]

Agreed that the latter is better. Where are we using just sizes?

Here are some of the runtimes using just sizes:

Halide/src/runtime/cuda.cpp

Line 1124 in d031408

size_t arg_sizes[],

Halide/src/runtime/opencl.cpp

Line 1077 in d031408

size_t arg_sizes[],

Halide/src/runtime/metal.cpp

Line 733 in d031408

size_t arg_sizes[],