metal : use autoreleasepool to avoid memory leaks (ggml-org#5437)

There appears to be a known memory leak when using the
`MLTCommandBuffer`. It is suggested to use `@autoreleasepool` in
[1,2]

[1] https://developer.apple.com/forums/thread/662721
[2] https://forums.developer.apple.com/forums/thread/120931

This change-set wraps the `ggml_metal_graph_compute` in a
`@autoreleasepool`.

This commit addresses ggml-org#5436

sync : ggml

server : fix prompt caching for repeated prompts (ggml-org#5420)

llama : do not cap thread count when MoE on CPU (ggml-org#5419)

* Not capping thread count when MoE inference is running on CPU

* Whitespace

ggml : fix `error C2078: too many initializers` for MSVC ARM64 (ggml-…

…org#5404)

Fix Vulkan crash on APUs with very little device memory (ggml-org#5424)

* Fix Vulkan crash on APUs with very little device memory

* Fix debug output function names

CUDA: more warps for mmvq on NVIDIA (ggml-org#5394)

llama : do not print "offloading layers" message in CPU-only builds (g…

…gml-org#5416)

Fix f16_sycl cpy call from Arc (ggml-org#5411)

* fix f16_sycl cpy call

* rm old logic

* add fp16 build CI

* use macro

* format fix

fix trailing whitespace (ggml-org#5407)