DAOS-16935 API: Add GPU direct I/O support

[gnailzenh]

Extend DAOS to support GPU direct RDMA I/O without changing the wire protocol or d_iov_t ABI. Key changes:

API layer:

• Add daos_mem_type_t enum (HOST, CUDA, CUDA_MANAGED, ROCM, ZE)
• Add daos_mem_attr_t side-channel struct for memory attributes
• Add DAOS_OBJ_IO_GPU_DIRECT flag for existing fetch/update APIs
• Add daos_obj_fetch_gpu()/daos_obj_update_gpu() wrappers

CaRT transport:

• Add crt_bulk_create_with_mem_attr() for GPU-aware bulk handles
• Forward memory type to Mercury via hg_bulk_attr
• Add D_GPU_DIRECT env var and crt_mem_device_enabled() init hook

Client object layer:

• Validate GPU direct flag against mem_attrs
• Propagate ORF_GPU_DIRECT to server RPCs
• Use crt_bulk_create_with_mem_attr() for GPU buffers

Server object layer:

• Add GPU direct observability (debug logs, telemetry counter)
• No behavioral change (Mercury handles GPU RDMA transparently)

Build system:

• Add BUILD_GPU_DIRECT SCons option (off by default)
• Conditionally enable CUDA/GDRCopy in UCX and FI_HMEM in libfabric

Steps for the author:

• Commit message follows the guidelines.
• Appropriate Features or Test-tag pragmas were used.
• Appropriate Functional Test Stages were run.
• At least two positive code reviews including at least one code owner from each category referenced in the PR.
• Testing is complete. If necessary, forced-landing label added and a reason added in a comment.

After all prior steps are complete:

• Gatekeeper requested (daos-gatekeeper added as a reviewer).

[github-actions]

Ticket title is 'dfs_write bad address from 32K buffer allocated with malloc_shared'
Status is 'In Progress'
Labels: 'ALCF,alcf_cluster'
Errors are Component should be lower-case
https://daosio.atlassian.net/browse/DAOS-16935

[daosbuild3]

Test stage NLT completed with status UNSTABLE. https://jenkins-3.daos.hpc.amslabs.hpecorp.net/job/daos-stack/job/daos//view/change-requests/job/PR-18419/1/testReport/