Remove Legacy Copy-OP Pointer Indirection Code

[anavp-nvidia]

As discussed in PR #16471, this PR removes the legacy copy-op pointer indirection code. This change allows cudaMemcpyAsync to be used instead of CUDA copy kernel for contiguous F32 tensors, resulting in ~4% performance improvement for Nemotron Nano v2 (NemotronH) model on RTX 5090.

Results:

Weights: bartowski/nvidia_NVIDIA-Nemotron-Nano-9B-v2-GGUF
Quantization: Q4_K_M

Performance before:

  Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| nemotron_h 9B Q4_K - Medium    |   6.07 GiB |     8.89 B | CUDA       |  99 |  1 |    tg200 @ d100 |        165.50 ± 0.19 |
| nemotron_h 9B Q4_K - Medium    |   6.07 GiB |     8.89 B | CUDA       |  99 |  1 | pp100+tg200 @ d100 |        174.14 ± 2.02 |

Performance after:

  Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| nemotron_h 9B Q4_K - Medium    |   6.07 GiB |     8.89 B | CUDA       |  99 |  1 |    tg200 @ d100 |        172.22 ± 0.16 |
| nemotron_h 9B Q4_K - Medium    |   6.07 GiB |     8.89 B | CUDA       |  99 |  1 | pp100+tg200 @ d100 |        180.91 ± 0.15 |

[slaren]

Since changing addresses of cpy operations in CUDA graphs is no longer supported, the exception for GGML_OP_CPY in ggml_graph_node_has_matching_properties should also be removed.

The indirections in cpy ops should also be removed, since their only purpose was to allow this, as well as ggml_cuda_cpy_dest_ptrs_copy and ggml_cuda_graph::cpy_dest_ptrs.