api!: rename the public CCCL_VERSION family to HIPCCL_*

[obersteiner]

The version macros exposed by libhipcxx carried NVIDIA's CCCL branding, which is misleading for consumers of a HIP library: CCCL_VERSION named libhipcxx's own release number, not any NVIDIA component's.

Rename the whole public family so the names match the project:

CCCL_VERSION -> HIPCCL_VERSION
CCCL_MAJOR_VERSION -> HIPCCL_MAJOR_VERSION
CCCL_MINOR_VERSION -> HIPCCL_MINOR_VERSION
CCCL_PATCH_VERSION -> HIPCCL_PATCH_VERSION

The value is unchanged (3000002) and the documented _LIBCUDACXX_CUDA_API_VERSION alias layer in <cuda/std/detail/__config> keeps working, so code using the documented spelling is unaffected.

Untouched on purpose: the CCCL_VERSION* macros in __cccl/compiler.h and __cccl/cuda_toolkit.h. Despite the shared spelling they are unrelated preprocessor machinery behind _CCCL_COMPILER(), CCCL_CUDACC() and CCCL_CTK(), not the release number.

Also:

• Point the libhipcxx-config-version.cmake regex at the new name. It scrapes the header as text, so nothing links it to the header; a stale pattern used to surface as a cryptic "math cannot parse the expression" during a downstream find_package(). Added an explicit FATAL_ERROR when the parse yields nothing.
• Guard the version test against the old spellings reappearing.

BREAKING CHANGE: CCCL_VERSION and its MAJOR/MINOR/PATCH companions are no longer defined. Because an undefined macro evaluates to 0 inside #if, code written as #if CCCL_VERSION >= ... will silently stop matching rather than fail to compile. Consumers should switch to HIPCCL_VERSION or to _LIBCUDACXX_CUDA_API_VERSION.

Description

closes

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[monoatamd]

LGTM, Thanks!