Add special value tests for reflect

Nobody has claimed this yet.

This issue tracks adding a test for special values like NaN, inf, and denormals, for the `reflect` function.
The behavior doesn't seem to have any consistency across different hardware, so part of the issue is figuring out how to standardize this behavior across different hardware so that the same results are seen. For example, one discrepancy is what to expect in this case:
`reflect({inf, 0.0}, {0.0, 1.0})`
In some environments, the result is expected to be {inf, 0.0}
In others, the result is {nan, nan}
This is the case for 32 bit floats, I presume 16 bit floats also need to be standardized.

No file or test path is named. Start by locating the existing reflect implementation and its tests in the offload-test-suite, then compare handling of NaN, infinity, and denormals across hardware. Done means the expected results are standardized for 32-bit and 16-bit floats, including reflect({inf, 0.0}, {0.0, 1.0}).

Written by the indexing model from the issue text.