feat: record variance partition and gp kernel fit in the manifest

[mohossam01]

What this PR does

Records two shape-characterisation summaries on every generated manifest. A per-metric variance partition quantifies how much of each metric's spread comes from the entity's archetype, from entity-level idiosyncrasy, and from within-entity time-series noise. A per-archetype Gaussian-process fit captures each trajectory's smoothness via a recovered length scale — short for oscillating shapes, long for monotone or gradual ones. Consumers can read both straight off the manifest to compare a downstream model's predictions against the engine's ground-truth statistics, with no re-generation and no separate fitting pipeline.

Files Modified

docs/site/api-reference.md
docs/site/manifest-reference.md
docs/site/user-guide/archetypes.md
plotsim/gp.py
plotsim/manifest.py
tests/test_manifest_variance_gp.py

How to test

python -m pytest tests/test_manifest_variance_gp.py
python -m pytest
python -m mypy plotsim/manifest.py plotsim/gp.py tests/test_manifest_variance_gp.py
python -m mkdocs build --strict