Add explicit layer argument for counterfactual/perturbation spatial…

[dbdimitrov]

… features (v1.1.2)

The training spatial_x is built from a normalized representation (log1p of CP10K) and adata.X is then reset to raw counts, because the model's NB likelihood needs counts. make_counterfactual_adata(precomputed=False) recomputed the counterfactual features with compute_spatial_features(adata), which reads adata.X — so edge-perturbation counterfactuals aggregated raw counts while the model had only ever seen log1p(CP10K) neighbourhood means. That train/inference mismatch silently degraded every precomputed=False counterfactual. make_neighbor_perturbation had the same issue on its input side: it wrote a perturbed layer and aggregated it, but _make_perturbed_expression always read adata.X as the source.

Rather than hard-coding a log1p inside the model module (which would guess at the caller's normalization), this adds an explicit layer argument:

• make_counterfactual_adata(..., layer=None) passes layer to compute_spatial_features when precomputed=False, threaded through Cellina._make_counterfactual_adata, Cellina.get_counterfactual_latents and Cellina.get_counterfactual_expression.
• make_neighbor_perturbation, _make_perturbed_expression and make_perturbed_expression take an optional source layer and read adata.layers[layer] instead of adata.X when given.
• compute_spatial_features warns when layer is None and adata.X looks like raw counts (sampled non-zero entries integer-valued with max > 50).

All new arguments default to None, so existing calls keep their behaviour. docs/tutorial.ipynb now stores adata.layers['lognorm'] before resetting X to counts and passes layer='lognorm' to both perturbation sections.

[codecov]

Codecov Report

❌ Patch coverage is 85.71429% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.06%. Comparing base (c77b614) to head (fc85cbc).

Files with missing lines	Patch %	Lines
src/cellina/_spatial_utils.py	85.71%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #41      +/-   ##
==========================================
- Coverage   90.10%   90.06%   -0.04%     
==========================================
  Files          10       10              
  Lines        1394     1419      +25     
==========================================
+ Hits         1256     1278      +22     
- Misses        138      141       +3     

☔ View full report in Codecov by Harness.
📢 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.