copy/ascend: add h2d ffts pipeline cases

[NaganooMei]

Summary

• Add optional Ascend H2D FFTS pipeline copy cases:
  • host_to_device_ffts_pipeline
  • one_host_to_all_device_ffts_pipeline
  • all_host_to_all_device_ffts_pipeline
• Add shared-host fan-out cases for multi-device reads:
  • one_share_host_to_all_device_ce
  • one_share_host_to_all_device_ce_multi_stream
  • one_share_host_to_all_device_ffts_pipeline
• Change Ascend multi-device all-host cases to fork one child per device for concurrent submit.
• Add forked result collection that merges child timing samples by per-iteration max cost and reports aggregate count/bandwidth.
• Add FragmentedDeviceCopyBuffer, isolated ascend/h2d_ffts_pipeline implementation, and FFTS SDMA dispatcher.
• Update readme.md with shared-host, multi-stream, fork-submit, and FFTS pipeline cases.

Scope

• This PR is limited to Ascend copy benchmark cases and the common runtime/result plumbing needed by those cases.
• one_host_to_all_device_* remains the ordinary aclrtMallocHost host0 path.
• one_share_host_to_all_device_* is the POSIX shared-memory path.
• Fork-submit cases skip parent ACL runtime initialization and create runtime state inside each child process.

Validation

COPY_FFTS_VALIDATE=0 COPY_FFTS_PIPELINE_OBJECT_FRAGS=64 ./build/module/copy/copy -t all_host_to_all_device_ffts_pipeline -s 32K -n 1024 -i 128 -d 8
COPY_FFTS_VALIDATE=0 COPY_FFTS_PIPELINE_OBJECT_FRAGS=64 ./build/module/copy/copy -t one_host_to_all_device_ffts_pipeline -s 32K -n 1024 -i 128 -d 8
COPY_FFTS_VALIDATE=0 COPY_FFTS_PIPELINE_OBJECT_FRAGS=64 ./build/module/copy/copy -t host_to_device_ffts_pipeline -s 32K -n 1024 -i 128 -d 8

Observed:

case	Copy avg	BW
all_host_to_all_device_ffts_pipeline	1912 us	130.753 GB/s
one_host_to_all_device_ffts_pipeline	2929 us	85.353 GB/s
host_to_device_ffts_pipeline	1395-1575 us per device	19.841-22.401 GB/s per device