Optimize comprehension evaluation with generator-based approach in LocalPythonExecutor

[albertvillanova]

Optimize comprehension evaluation with generator-based approach in LocalPythonExecutor.

This PR refactors the comprehension evaluation logic to use a generator-based approach for better performance and memory efficiency.

Performance Benefits

1. Memory efficiency: Eliminates intermediate list allocations during comprehension evaluation
2. Lazy evaluation: Only computes what's needed, especially beneficial for set() and dict() where duplicates are discarded
3. Optimized construction: list(generator) and set(generator) are highly optimized in CPython
4. Reduced overhead: No more list wrapping/unwrapping in base case evaluators

Memory Complexity Improvement

The refactored implementation reduces memory overhead from O(n) to O(1) for the iteration mechanism itself:

• Before: Built complete intermediate lists at each recursion level, requiring O(n) memory allocation per level
• After: Yields elements one at a time using generator state, maintaining only O(1) memory for the iteration machinery regardless of comprehension size

Follow-up to:

• LocalPythonExecutor: Add support for nested dictcomp and setcomp #1823