Add missing CUDA repo rule features to bzlmod

[ybaturina]

Bazel workspace mode allowed to pass custom redistributions and build templates to the repository rules.
Now this feature is added to bzlmod too.

Example of MODULE.bazel configuration:

CCCL_DIST_DICT = {
 "cuda_cccl;linux-sbsa;full_path": "\"https://github.com/NVIDIA/cccl/releases/download/v2.8.5/cccl-src-v2.8.5.tar.gz\"",
 "cuda_cccl;linux-sbsa;sha256": "\"631cb620e78904fb30f4ef6e033efb7716af1934bc170dce9cbbf4cc48777ca2\"",
 "cuda_cccl;linux-sbsa;strip_prefix": "\"cccl-src-v2.8.5\"",
 "cuda_cccl;linux-x86_64;full_path": "\"https://github.com/NVIDIA/cccl/releases/download/v2.8.5/cccl-src-v2.8.5.tar.gz\"",
 "cuda_cccl;linux-x86_64;sha256": "\"631cb620e78904fb30f4ef6e033efb7716af1934bc170dce9cbbf4cc48777ca2\"",
 "cuda_cccl;linux-x86_64;strip_prefix": "\"cccl-src-v2.8.5\""
}
CCCL_GITHUB_VERSIONS_TO_BUILD_TEMPLATES = {
  "cuda_cccl;repo_name": "\"cuda_cccl\"",
  "cuda_cccl;local;source_dirs": "[\"include\",\"lib\"]",
  "cuda_cccl;version_to_template;any": "\"@rules_ml_toolchain//gpu/cuda/build_templates:cuda_cccl_github.BUILD.tpl\"",
  "cuda_cccl;local;version_to_template;any": "\"@rules_ml_toolchain//gpu/cuda/build_templates:cuda_cccl.BUILD.tpl\""
}

cuda_json_ext = use_extension("@rules_ml_toolchain//extensions:cuda_json.bzl", "cuda_json_ext")
use_repo(cuda_json_ext, "cuda_redist_json")

cuda_redist_init_ext = use_extension("@rules_ml_toolchain//extensions:cuda_redist_init.bzl", "cuda_redist_init_ext")
cuda_redist_init_ext.configure(
    cuda_redistributions = CCCL_DIST_DICT,
    redist_versions_to_build_templates = CCCL_GITHUB_VERSIONS_TO_BUILD_TEMPLATES,
)
use_repo(
    cuda_redist_init_ext,
    "cuda_cccl",
    "cuda_crt",
    "cuda_cublas",
    "cuda_cudart",
    "cuda_cudnn",
    "cuda_cufft",
    "cuda_cupti",
    "cuda_curand",
    "cuda_cusolver",
    "cuda_cusparse",
    "cuda_driver",
    "cuda_nvcc",
    "cuda_nvrtc",
    "cuda_nvdisasm",
    "cuda_nvjitlink",
    "cuda_nvml",
    "cuda_nvtx",
    "cuda_nvvm",
    "cuda_profiler_api",
)

cuda_configure_ext = use_extension("@rules_ml_toolchain//extensions:cuda_configure.bzl", "cuda_configure_ext")
use_repo(cuda_configure_ext, "local_config_cuda")

nvshmem_json_ext = use_extension("@rules_ml_toolchain//extensions:nvshmem_json.bzl", "nvshmem_json_ext")
use_repo(nvshmem_json_ext, "nvshmem_redist_json")

nvshmem_redist_init_ext = use_extension("@rules_ml_toolchain//extensions:nvshmem_redist_init.bzl", "nvshmem_redist_init_ext")
use_repo(
    nvshmem_redist_init_ext,
    "nvidia_nvshmem",
)

nccl_redist = use_extension("@rules_ml_toolchain//extensions:nccl_redist.bzl", "nccl_redist_ext")
use_repo(nccl_redist, "cuda_nccl")

nccl_configure = use_extension("@rules_ml_toolchain//extensions:nccl_configure.bzl", "nccl_configure_ext")
use_repo(nccl_configure, "local_config_nccl")

CCCL_DIST_DICT and CCCL_GITHUB_VERSIONS_TO_BUILD_TEMPLATES are nested dictionaries that were processed by flatten_dict_for_string_dict defined in in //gpu:nvidia_common_rules.bzl.

[yuriivcs]

LGTM