Kernel overlap verification stage, initial version

[lasch]

introducing initial version of verification stage for kernel overlap detection.

Main functionality for overlap verification: src/aiu_trace_analyzer/verification/overlap_verify.py

On top of the derived context, a few tweaks were needed to support the requirements:

• strict mode: needed to flag regular function call (fully nested overlaps) as actual overlaps
• more flexible hash-keys: hard-coded pid+tid needs to be more configurable to cover actual 'streams'
  • new implementation uses default steam 0 if no args.stream is in the event
  • this will cause memory events to get treated within their own stream until Enable stream id for mem events torch-spyre/torch-spyre#3699 is fixed
• issue_warning required an override-capable extension to allow derived classes to issue 'warning instances' for the verification report

Also included the following cleanup:

• verification functions/context imports moved to their own __init__.py
• removed the template/example verify stage from the verification pipeline
• verification pipeline needed global sorting stage

@nicksunka @ppnaik1890 I did not yet scrutinize the event filter - right now, any event that comes with indication of 'ran on accelerator' is being checked for overlap within its stream.

[ppnaik1890]

minor clarification questions

[lasch]

@ppnaik1890 are the provided explanations sufficient?