Profile layer: Add profile layer

[solidpixel]

This PR adds a profiling layer which can collect API-correlated performance information, such as performance counters, from Arm GPUs.

Task list

• Add CPU-GPU sync points for all workloads
• Add pipeline barrier serialization for all workloads (see support layer)
• Add queue submit serialization for all workloads (see support layer)
• Add config handling to enable only for selected frames
• Filter sync and serialization so it only applies to the selected ranges
• Update developer documentation to reflect final design
• Update user documentation to reflect final design

For future PRs:

• Add config handling to enable based on times not frames.
• Add custom counter selection to the config files.
• Add timeline-like metadata export into the CSV files.
• Make additional time wait conditional (although seems low value).
• Expose debug_utils even if the driver doesn't (see Allow a layer to implement extensions the driver doesn't #134)