Issue 51: Handle flat targets for GP fit

[SamuelBrand1]

This PR introduces two fixes for adding robustness for near-constant targets so AutoGP fits don't produce singular covariances.

1. _stabilize_for_fit(y; flat_threshold=1e-3) which adds small Gaussian jitter when the transformed series' relative range is below the threshold, and uses it in make_and_fit_model to construct the GP with stabilized input data.
2. In transformations, detect pathological Box-Cox fits that collapse the transformed range and fall back to the positive (log) transformation with a warning.

Add tests covering Box-Cox fallback, make_and_fit_model behavior on near-constant and exactly constant data, and unit tests for _stabilize_for_fit behavior. This prevents PosDefException during fitting/forecasting (issue #51) while preserving existing behavior for healthy data.

Pinging @damonbayer for awareness.

[Copilot]

Pull request overview

This PR addresses issue #51 where AutoGP fits raised PosDefException on near-constant target series (e.g., flat hospital admission counts where the Box-Cox MLE selects a pathological λ that collapses the transformed range to a constant, yielding a singular GP covariance). Two complementary safeguards are introduced, plus tests.

Changes:

• Add _stabilize_for_fit in make_and_fit_model.jl that injects small Gaussian jitter when the (transformed) target's relative range falls below flat_threshold (default 1e-3), and wires it into make_and_fit_model via a new flat_threshold kwarg.
• In get_transformations("boxcox", ...), evaluate the fitted Box-Cox transform on the data and fall back to the "positive" (log) transformation with a @warn if the transformed range is non-finite or collapses to ≤ 1% of the log range.
• Add regression tests for Box-Cox fallback, flat/exactly-constant data end-to-end fits + forecasts, and unit tests for _stabilize_for_fit.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/make_and_fit_model.jl	New _stabilize_for_fit helper and flat_threshold kwarg; applies jitter to data.y before constructing AutoGP.GPModel.
src/transformations.jl	Detect degenerate Box-Cox fits and fall back to log transformation with warning; updated docstring.
test/test_helper_functions.jl	Test for Box-Cox fallback on near-constant data and that healthy data still uses Box-Cox.
test/test_model_fitting.jl	Tests for end-to-end fit+forecast on near-constant and exactly-constant data, and unit tests for _stabilize_for_fit.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.68%. Comparing base (c5549c4) to head (bd265f4).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #74      +/-   ##
==========================================
+ Coverage   95.08%   95.68%   +0.60%     
==========================================
  Files           5        5              
  Lines         122      139      +17     
==========================================
+ Hits          116      133      +17     
  Misses          6        6              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.