Make forecast boosting rounds configurable

[amarakramali]

What changed

• add --estimators to the production forecast CLI
• forward the selected count to final XGBoost training
• document the option in the configurable forecast example
• add a command-level regression test covering option forwarding and outputs

Why

The evaluator and demo already allowed the number of boosting rounds to be
configured, and the README said this setting was configurable. The production
aep-forecast command nevertheless always used the hard-coded default of 800.

User impact

Users can now trade training time against model capacity consistently across
evaluation, demo, and production forecast commands.

Validation

• python -m pytest -q (66 passed)
• python -m ruff check src tests streamlit_app.py
• python -m compileall -q src tests streamlit_app.py
• python -m build
• verified installed aep-forecast --help exposes --estimators