Test `clSetKernelArg` preserves `cl_kernel` state on most error codes

Assuming that this specification clarification merges (https://github.com/KhronosGroup/OpenCL-Docs/pull/705) there could be CTS tests that deliberately cause `clSetKernelArg` to return an error code but ensure that `cl_kernel` state is unmodified.

This could be done by setting a kernel argument correctly, try to overwrite that kernel argument with an input that will result in an error code, then check that the original argument was preserved via `clGetKernelArgInfo`.

Invalid inputs could be used to trigger any or all of these error codes:

- `CL_INVALID_KERNEL`
- `CL_INVALID_ARG_INDEX`
- `CL_INVALID_ARG_VALUE`
- `CL_INVALID_MEM_OBJECT`
- `CL_INVALID_SAMPLER`
- `CL_INVALID_DEVICE_QUEUE`
- `CL_INVALID_ARG_SIZE`
- `CL_MAX_SIZE_RESTRICTION_EXCEEDED`
- `CL_INVALID_ARG_VALUE`

Start by reading the linked OpenCL-Docs specification clarification and locating existing OpenCL-CTS coverage for clSetKernelArg, clGetKernelArgInfo, and the listed error codes. Add cases that set an argument successfully, provoke each applicable error, and verify the original argument remains unchanged; done means the new CTS tests pass.

Written by the indexing model from the issue text.