Amd trace lifecycle

[CodingInAVan]

Description

This hardens the AMD ROCprofiler backend across tracing, deep-window profiling, and device attribution.

• Flush ROCprofiler activity at segment boundaries so completed records remain in the correct segment.
• Track and report buffer flush failures, client queue drops, and scope-correlation failures.
• Preserve scope push/pop correctness across backend lifecycles.
• Implement AMD deep-window readiness, servicing, arming, and engine reporting.
• Honor DeepArmMode::WindowOnly for dispatch counting.
• Enforce launch budgets atomically without running teardown from the dispatch callback.
• Attribute dispatch-counter samples using ROCprofiler agent mappings.
• Reject dispatch-counter configuration and records from unconfigured or unknown GPU agents instead of mislabeling them as device 0.

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