Make baseline evaluation configurable

[amarakramali]

Summary

• make the baseline evaluator import-safe and configurable through CLI arguments
• validate timestamps, hourly continuity, and required numeric feature columns
• support custom metrics and plot destinations while creating parent directories
• document baseline evaluation in the synthetic no-download quickstart
• cover metrics, invalid inputs, and generated command artifacts with tests

Why

The synthetic quickstart could generate feature data, but baseline evaluation still executed on import and used hard-coded production paths. That prevented users from evaluating the bundled sample workflow or safely reusing the evaluation functions in tests and other code.

Impact

Existing default paths and the 30-day evaluation window remain available. Users can now evaluate any compatible feature CSV, choose evaluation and plot windows, and place artifacts wherever needed.

Validation

• python -m pytest -q — 56 passed
• uvx --from ruff==0.16.0 ruff check src tests streamlit_app.py — passed
• python -m compileall -q src tests streamlit_app.py — passed
• synthetic 40-day data → features → 30-day baseline evaluation smoke test — metrics CSV and 101 KB PNG generated