Fix heapUsage query for non-unified memory devices

[SRSaunders]

Fixes #2082 for non-unified memory machines. No change for unified memory machines.

1. Replaces task_vm_info.phys_footprint with task_vm_info.ledger_tag_graphics_footprint in mvkGetUsedMemorySize(). This provides a more accurate estimate of mapped memory for heapUsage[1] on non-unified memory machines. As an aside, it also provides a more accurate fall-back within MVKPhysicalDevice::getCurrentAllocatedSize() if the selector is not available for any machine.
2. Fixes the calculation for heapUsge[0] on non-unified memory machines, where it represents the device-only memory. The calculation remains the same for unified memory machines (where heapUsage[1] = 0).

The only edge case I see is for non-unified memory machines where the currentAllocatedSize selector is not available. In this case heapUsage[0] will return 0 for device memory and heapUsage[1] will return mapped memory. I think this is a reasonable fallback if the total memory information is not available from Metal.

[CLAassistant]

All committers have signed the CLA.

[SRSaunders]

One little change while you merge:

uint64_t currentAllocatedSize should be declared either auto or VkDeviceSize.

Sorry for missing this.