[FEA] Refactor AutoTuner into separate classes for Qualification and Profiling Tools

[parthosa]

Description
In #1397, we identified that many AutoTuner recommendations need to differ based on the context in which it is run:

• When AutoTuner processes GPU event logs (via the Profiling Tool), the recommendations align in improving the GPU based runs.
• When AutoTuner processes CPU event logs (via the Qualification Tool), different recommendations are required for transition to the first-GPU run.

Currently, there is no clean separation between these two case.

Proposed Solution

• Introduce a class-based design to create specialized AutoTuner implementations for Qualification.
• QualificationAutoTuner subclass will override specific configurations or logic as needed.