Added warning messages for debugging

[devalshahamd]

This pull request introduces new diagnostic features to improve performance analysis and debugging in the PyTorch inference report generation and trace merging processes. The main changes add automated warnings for potential performance issues related to synthetic and uncategorized operations, and enhance failure reporting during graph capture merges.

Performance diagnostics and reporting improvements:

• Added the diag_sweep_unified_perf_summary function to generate_perf_report_pytorch_inference.py, which scans the unified performance summary and prints diagnostic warnings if synthetic or uncategorized operations exceed a configurable runtime threshold. This helps users quickly identify potential inefficiencies.
• Integrated the new diagnostic sweep into the generate_perf_report_pytorch workflow, ensuring that these warnings are automatically generated as part of the performance report.

Graph capture merge diagnostics:

• Modified the merge loop in merge_capture_trace_into_graph to use an explicit step counter, enabling more informative diagnostics during processing.
• Enhanced failure reporting in merge_capture_trace_into_graph by printing a detailed diagnostic message (DIAG) when graph capture merging fails, including step number and event/root counts for both capture and graph traces. This provides clearer context for debugging merge failures.