Metal Pool 1D Kernel

[ThoreKoritzius]

This PR adds Metal backend support for POOL_1D operators.
Until now, only the POOL_2D was implemented for Metal.
With this change, models relying on 1D pooling can now run fully accelerated on Apple GPUs through Metal in an analog way.

---

Implementation Details

• Added new Metal compute kernels in ggml-metal.metal:

  • kernel_pool_1d_max_f32
  • kernel_pool_1d_avg_f32

• Implemented ggml_metal_op_pool_1d analogous to ggml_metal_op_pool_2d, including:

  • Argument struct (ggml_metal_kargs_pool_1d)x
  • Threadgroup size calculation
  • Encoding input/output buffers

[ggerganov]

• Need to add tests
• The implementation assumes contiguous data - either require this to be true, or extend the implementation to support non-cont data

[ggerganov]

The order of the arguments does not match the initialization in ggml-metal-ops.cpp.

Also, some of these arguments like np need to be 64-bit integers.

[tommarques56]

Hi @ThoreKoritzius, I can run an automated high-severity-only LLM review on this PR and post a single focused inline comment. Reply with "approve" or add a comment saying "@tommarques56 approve" to proceed.