Make math_builtin_api tests splittable into smaller chunks

[jzc]

Currently, the some of the generated math_builtin_api tests are huge. These are the current line counts:

   72405 tests/math_builtin_api/math_builtin_integer.cpp
   27088 tests/math_builtin_api/math_builtin_float_half.cpp
   27088 tests/math_builtin_api/math_builtin_float_double.cpp
   27079 tests/math_builtin_api/math_builtin_float_base.cpp
   17218 tests/math_builtin_api/math_builtin_relational_base.cpp
    4599 tests/math_builtin_api/math_builtin_relational_half.cpp
    4599 tests/math_builtin_api/math_builtin_relational_double.cpp
    4307 tests/math_builtin_api/math_builtin_common_half.cpp
    4307 tests/math_builtin_api/math_builtin_common_double.cpp
    4298 tests/math_builtin_api/math_builtin_common_base.cpp
    3689 tests/math_builtin_api/math_builtin_native.cpp
    3689 tests/math_builtin_api/math_builtin_half.cpp
    1163 tests/math_builtin_api/math_builtin_geometric_base.cpp
     832 tests/math_builtin_api/math_builtin_geometric_double.cpp

When building on release mode, -O3 optimization will be used, which aside from making compilation taking much longer, will make building math_builtin_integer.cpp use an unreasonable amount of RAM for the average person (64GB+).

This PR adds a functionality to split these tests into smaller chunks, which allows for each chunk to be compiled with a reasonable amount of RAM, and leads to a much faster compilation time.

Each of the original generated files listed above are all split into a test-wide configurable amount SYCL_CTS_MATH_BUILTIN_N_SPLITS, which is 16 initially. This makes each of the chunks of the biggest test math_builtin_integer.cpp under 5k LOC.

[AlexeySachkov]

I guess that this if means that geometric_half is small enough by itself to be split, but a comment would be helpful here about this exception

[AlexeySachkov]

Why do we need SYCL_CTS_MATH_BUILTIN_N_SPLITS in the name?
If we decide to preserve it, would it be better to add _of_ between split index and total number of splits to make the name more human-readable?

[keryell]

@jzc @steffenlarsen How does this compare to #927?

[steffenlarsen]

@jzc @steffenlarsen How does this compare to #927?

I did not even realize @jzc had a patch like this too. They look very similar, so I feel bad for intercepting it with my patch. Sorry, @jzc !

[keryell]

@jzc Are there more improvements to do on top of #927?

[jzc]

@keryell No, Steffen's patch should cover it, closing this PR.