GPU Metal: accept the registration of host memory

The GPU reconstruction workflow registers its shared-memory segments
for the GPU, and the Metal backend fell back to the device base, which
fails for every GPU backend that does not implement it: the full
system test's WITHGPU stage stopped with "Error registering memory
for GPU". Metal memory is shared with the host, and GPUMemCpy copies
from memory outside the Metal buffers on the host, so there is nothing
to pin: accept the registration as the CPU backend does.

Author: ktf

GPU/GPUTracking/Base/metal/GPUReconstructionMetal.h

@@ -46,6 +46,10 @@ class GPUReconstructionMetal : public GPUReconstructionProcessing::KernelInterfa
   size_t GPUMemCpy(void* dst, const void* src, size_t size, int32_t stream, int32_t toGPU, deviceEvent* ev = nullptr, deviceEvent* evList = nullptr, int32_t nEvents = 1) override;
   void ReleaseEvent(deviceEvent ev) override;
   void RecordMarker(deviceEvent* ev, int32_t stream) override;
+  // The memory is shared with the host, and GPUMemCpy copies from memory outside the Metal buffers on the
+  // host, so there is nothing to pin.
+  int32_t registerMemoryForGPU_internal(const void* ptr, size_t size) override { return 0; }
+  int32_t unregisterMemoryForGPU_internal(const void* ptr) override { return 0; }
 
   template <class T, int32_t I = 0>
   int32_t AddKernel();