feat: record per-pair regression and seasonal lift in the manifest

[mohossam01]

What this PR does

Writes regression and seasonal modulation summaries into every generated manifest. Anyone analyzing a plotsim output dataset can now read the linear relationship between each pair of correlated metrics — both overall and within each archetype — and compare a fitted model against ground truth without re-running generation. The seasonal numbers let a consumer subtract the engine's periodic lift from any cell before scoring.

Files Modified

plotsim/tables.py
plotsim/manifest.py
plotsim/cli.py
tests/test_manifest_decomposition_regression.py
tests/test_manifest.py
tests/test_heteroscedastic_noise.py
tests/test_multi_source.py
tests/test_per_metric_treatment.py
tests/test_time_varying_correlations.py
CHANGELOG.md
docs/site/manifest-reference.md
docs/site/api-reference.md
docs/site/user-guide/seasonality.md
docs/site/user-guide/metrics-and-connections.md

How to test

python -m pytest tests/test_manifest_decomposition_regression.py
python -m pytest tests/test_manifest.py tests/test_heteroscedastic_noise.py tests/test_multi_source.py tests/test_per_metric_treatment.py tests/test_time_varying_correlations.py
python -m pytest
python -m mkdocs build --strict