Add new gpu_timeline layer

[solidpixel]

This PR adds a new layer used for exporting semantic information about running workloads for Arm GPU timeline visualization.

Task list

The task list for this layer is shown below. Where possible we'll split these off into smaller PRs to make merges more manageable and then rebase this PR on top.

• Implement network comms module for sending data to the host.
• Implement prototype server to capture data on the host.
• Enable VK_EXT_debug_utils extension for debug markers if app hasn't enabled it already.
• Add metadata generation for transfer workloads.
• Add metadata generation tracking for compute workloads.
• Improve error reporting for devices that are not compatible.
• Emit transfer sizes in pixels/bytes.
• Correctly detect and emit not supported placeholder values for corner-case usage.
  • Not supported: Dynamic render passes from Vulkan 1.3 that are split across command buffers where the start piece is not one-time submit.
  • Indirect compute workloads have an unknown dispatch size.

To be handled in follow-on PRs:

• Track VkImage and VkBuffer resources so we can emit sizes for transfers.
• Improve the prototype Python host server to make it production ready.
• Add metadata protocol support for an arbitrary user string in addition to the label to allow users to expand the visualization in Streamline by modifying the layer.
• Optimize if performance is an issue
  • Currently we use JSON as it was sufficient for rapid prototyping - should be replaced with a simple binary packet encoding to avoid the memory allocation churn.