Fix precomputed embeddings not being wired through to the model

[mojtababahrami]

Summary

• api.py never passed precomp_embs_key into ContrastiveModel when constructing or loading the model, so self.precomp_embs_key stayed None and the model always fell back to computing embeddings from scratch instead of using precomputed ones.
• AnnDataModule passed obsm_keys=precomp_embs_key (a bare string) instead of wrapping it in a list, which is what the underlying dataset class expects.
• predict_step unconditionally normalized cell_embs, even when precomp_embs_key is set, which is inconsistent with _step's distance-based (unnormalized) logic for precomputed embeddings.

Test plan

• python -m py_compile on the changed files
• No new automated tests added (user opted to skip, since this is a targeted bug fix)

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.03%. Comparing base (a49b310) to head (c8d0b34).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #27      +/-   ##
==========================================
+ Coverage   66.99%   67.03%   +0.03%     
==========================================
  Files          22       22              
  Lines        3003     3006       +3     
==========================================
+ Hits         2012     2015       +3     
  Misses        991      991              

Files with missing lines	Coverage Δ
src/concept/api.py	76.71% <100.00%> (+0.04%)	⬆️
src/concept/data/datamodules.py	90.85% <100.00%> (+0.05%)	⬆️
src/concept/model.py	69.37% <100.00%> (+0.04%)	⬆️

🚀 New features to boost your workflow:

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