build: split reporting into optional extras

[s3alfisc]

Summary

• Reduce the base install to the packages needed for estimation and inference: Formulaic, joblib, Narwhals, NumPy, pandas, and SciPy.
• Add tables and plots extras, retain numba and torch, and add all as the union of those end-user feature extras.
• Give actionable installation errors such as pip install "pyfixest[tables]" when an optional feature is called.
• Add a minimal-install CI job that installs only the base package, fits a model, calls summary(), and verifies that optional packages were not installed.
• Regenerate pixi.lock for the dependency split.

Benchmark-only dependencies remain in the existing benchmarks extra and are intentionally excluded from all.

Validation

• Clean temporary environment installed the base project with only its six direct dependencies and passed an estimation/summary smoke test
• Verified all equals the union of tables, plots, numba, and torch
• pixi run test-py (complete stack: 805 passed, 13 skipped, 1 xfailed)
• Workflow validation, TOML checks, Ruff, and mypy

Stack

1. refactor: lazy-load reporting dependencies #1461 — lazy-load reporting dependencies
2. build: split reporting into optional extras #1462 — split reporting into optional extras
3. docs: explain optional dependency extras #1463 — explain optional dependency extras

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}