Amd sync current main

[CodingInAVan]

Description

This PR begins syncing the AMD ROCprofiler backend with the current client architecture.

It separates cross-vendor profiling intent from the actual AMD service selected at runtime and fixes multi-GPU device attribution for AMD trace records.

Changes

• Use ROCprofiler-native service names:

  • amd.buffer_tracing
  • amd.pc_sampling
  • amd.dispatch_counting
  • amd.device_counting

• Preserve vendor-neutral request intent such as sass_metrics and range_profiler.

• Report unsupported requests as explicit degraded fallbacks.

• Emit capture capability and data-quality information at shutdown.

• Correct device attribution for:

  • Kernel dispatches
  • Host-to-device copies
  • Device-to-host copies
  • Device-to-device copies

• Drop and report records with unmapped GPU agents instead of incorrectly assigning device 0.

• Add device_id as a backward-compatible trailing column in memcpy_event_batch.

• Update the Python analyzer to support both old and new memcpy batch formats.

• Reset ROCprofiler agent state safely between backend lifecycles.

• Add GPU-independent AMD policy and attribution tests.

Type of Change

• Bug fix
• New feature
• Documentation update

Testing

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas