Fix A100/H100 profiles, add examples + GitHub Pages docs, exclude toy_models from distribution, prep for PyPI

[vh2001]

• Hardware profile fixes: corrected nvidia_a100_80gb's FP16 throughput (312 → 77.97 TFLOPS) and TDP (300 → 400 W), and nvidia_h100_sxm's FP16 throughput (990 → 267.7 TFLOPS), in src/ecal/hardware/data/profiles.yaml. Updated the matching tables in README.md and docs/configuration.md to match.

• Examples: added examples/basic_estimate.py (minimal single-call ecal.estimate() usage) and examples/full_pipeline.py (the same computation walked through stage-by-stage via the low-level Transmission/DataPreprocessing/Training/Inference classes, useful as a reference for anyone needing finer control than estimate() exposes).

• GitHub Pages deployment for docs: added .github/workflows/docs.yml — builds the Sphinx docs with -W --keep-going (warnings fail the build) on every push/PR, and deploys to GitHub Pages on pushes to main or manual dispatch.

• Excluded ecal.calculators.toy_models from the installable package: it's experimental scratch code (untested, not exported, and depends on efficient_kan, which isn't a declared dependency). Its content now lives only in the legacy calculators/ToyModels.py, alongside scripts//notebooks/, rather than shipping inside the ecal package.

• PyPI publishing prep:

  • Removed the unused setuptools-scm build dependency — it wasn't driving versioning (version is static) but its presence made setuptools default to bundling the entire repo into the sdist (legacy dirs, notebooks, docs source, CI config, etc.).
  • Added MANIFEST.in so the sdist explicitly includes the full tests/ directory (conftest.py/__init__.py were otherwise silently dropped).
  • Added the author email and a Documentation project URL; updated Homepage to https://github.com/sensorlab/eCAL.