metal: add support for LOG op (f32, f16)

[RD-zhang1234]

Make sure to read the contributing guidelines before submitting a PR

Summary

Adds support for the LOG operation to the Metal backend for both f32 and f16 tensor types.

Implementation details

• Added four kernels in ggml-metal.metal:
  • kernel_log_f32
  • kernel_log_f32_4
  • kernel_log_f16
  • kernel_log_f16_4
• Updated ggml-metal-device.m to allow F16 for the LOG op.
• Integrated LOG into the unified unary op path (same as SIN and COS).

Validation

Tested on Apple M1 (MTLGPUFamilyApple7)

./build/bin/test-backend-ops -b Metal -o LOG
./build/bin/test-backend-ops perf -b Metal -o LOG
✅ 15978 / 15978 tests passed
✅ f32 / f16 variants all loaded successfully:
	•	kernel_log_f32
	•	kernel_log_f32_4
	•	kernel_log_f16
	•	kernel_log_f16_4
Notes
	•	f16 kernels compute in f32 for better numerical stability, then cast back to half.
	•	Matches CPU backend results for the natural logarithm (log(x)).
	•	Performance parity verified with test-backend-ops perf.