Split UR_RESULT_ADAPTER_SPECIFIC_ERROR into two separate ur_result_t, for warnings and for errors

[hdelan]

UR_RESULT_ADAPTER_SPECIFIC_ERROR can be used to emit adapter specific warnings as well as errors.

In CUDA adapter warnings are reported if setErrorMessage is called with UR_SUCCESS followed by a return of UR_RESULT_ADAPTER_SPECIFIC_ERROR. Errors are reported if setErrorMessage is called with UR_RESULT_ADAPTER_SPECIFIC_ERROR followed by a return of UR_RESULT_ADAPTER_SPECIFIC_ERROR. This is unintuitive. It would be nice if we could return either

• UR_RESULT_ADAPTER_SPECIFIC_ERROR for errors
• UR_RESULT_ADAPTER_SPECIFIC_WARNING for warnings

And setErrorMessage would not need to take an extra ur_result_t param.