refactor(ops): use InfiniOps Copy for rearrange

[voltjia]

What

• Replace RearrangeInfinilm with the canonical InfiniOps Copy API.
• Pass non_blocking=false, preserving the current synchronous copy behavior.
• Remove the obsolete adapter-blocking comment.

Alignment

InfiniCore wrapper	InfiniOps call	Alignment basis
rearrange(x, y)	Copy(input=x, non_blocking=false, out=y)	Tensor.copy_(src, non_blocking=False), InfiniOps Copy, InfiniOps #872, provider adaptation #897

InfiniOps keeps its C++ input/attribute/output ordering while retaining PyTorch's source tensor and non_blocking contract.

Why

The canonical Copy operator and CUDA adaptation are now available, so this adapter no longer needs the deprecated InfiniLM-specific rearrange class.

Scope

This is stacked on #1477. It changes only the InfiniOps adapter target; the public InfiniCore rearrange API and copy semantics remain unchanged.

Screenshots: N/A (backend adapter migration only).

Validation

Run on ssh nvidia in accelerator-dev/nvidia:latest on NVIDIA A100 GPUs:

• Full infinicore_cpp_api and _infinicore build/install passed.
• Python extension import and dynamic linking passed.
• Existing NVIDIA suites completed after the copy migration: sigmoid 45/45, SiLU-and-mul 36/36, and conv2d 12/12.
• python3 scripts/format.py --ref 850fb3f7 --path src --check with clang-format 16.0.6.
• git diff --check.

[voltjia]

Superseded by #1480, which consolidates the canonical InfiniOps adapter migrations while preserving each logical change as a separate commit.