[UR][CUDA] Mark MAX_READ_WRITE_IMAGE_ARGS as optional query
CUDA does not support read-write images in the same sense as OpenCL.
CUDA has separate concepts:
- Texture objects (read-only)
- Surface objects (read-only/write via surfaces)
These are distinct API types and cannot be used as a single 'read-write'
image argument to a kernel, which is what MAX_READ_WRITE_IMAGE_ARGS
queries.
Changes:
1. Mark MAX_READ_WRITE_IMAGE_ARGS as [optional-query] in device.yml spec
2. Add explanatory comment to device.yml describing why this is optional
3. Regenerate ur_api.h and optional_queries.h from spec
4. Remove xfails for CUDA and NativeCPU from SuccessMaxReadWriteImageArgs test
The CUDA adapter returns UNSUPPORTED_ENUMERATION for this query, which
is now correctly handled as an optional feature by the conformance test.
NativeCPU also returns UNSUPPORTED_ENUMERATION as it has no IL/image support.
Tests now pass for both adapters.
