Centralize output directory management

[yoonspark]

Currently, logic to manage pipeline output directory is split and duplicated between Pipeline and PipelineOutput. To avoid potential drift between these two, we want to centralize the relevant logic. Specifically, we could update PipelineOutput to be a single authoritative interface to create, validate, and interact with output directory structure; and have Pipeline to use PipelineOutput for operations related to output directory. PipelineOutput could be relocated to pipeline.py along with Pipeline (its current location in models.py is confusing anyway).

[yoonspark]

Note that tests for PipelineOutput are also scattered across multiple modules (i.e., test_report.py and test_models.py). They need to be centralized as well.

[yoonspark]

PipelineOutput is using several Pydantic models as its operational data stores, which is unnecessary and inefficient. We should replace them with more lightweight data classes.