Metal: Implement query pools

[squidbus]

Implements timestamp query pools for Metal using counter sample buffers.

• If an encoder is active and the required sampling point is supported for sampling in the current encoder, just encode a sampleCountersInBuffer in the active encoder.
• If an encoder is not active, or sampling the current encoder is not supported, and blit sampling is supported, use a new blit encoder to encode a sampleCountersInBuffer.
• In all other cases, end the current active encoder and use a temporary blit encoder to wait for a fence and perform stage-boundary sampling.

I've checked this with UnleashedRecomp and both the dependency graph and timestamps look similar to what I get from MoltenVK on an Apple GPU.

EDIT: Fixes #8.

[squidbus]

Rebased on latest main.

[DarioSamo]

Gonna wait on an approval from @dcvz to make sure this works as intended first.

[DarioSamo]

Needs a rebase.

[squidbus]

Rebased