Fix #10627: post_step clamping process doesn't preserve training labe...

[JiwaniZakir]

Closes #10627

The default post_step in LabelPropagation.forward (torch_geometric/nn/models/label_prop.py) clamped propagated values to [0, 1] but did not restore labeled node values afterward, causing training labels to be overwritten and collapsing class evidence at high-degree nodes. After the out.clamp_(0., 1.) call, labeled node outputs are now reset to their original one-hot values via out[mask] = y[mask] when mask is not None.

• torch_geometric/nn/models/label_prop.py — LabelPropagation.forward: added two lines after the default clamp to restore out[mask] from y[mask]
• test/nn/models/test_label_prop.py — test_label_prop: added a regression case using a star graph with 6 class-0 and 2 class-1 labeled neighbors aggregating into a single unlabeled hub node, asserting torch.allclose(out2[mask2], y2_oh[mask2].float())

The fix is verified by the new test, which previously failed because the labeled nodes' outputs diverged from their one-hot targets after clamping.

---

This PR was created with AI assistance (Claude). The changes were reviewed by quality gates and a critic model before submission.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.94%. Comparing base (c211214) to head (fbba04c).
⚠️ Report is 202 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #10656      +/-   ##
==========================================
- Coverage   86.11%   83.94%   -2.17%     
==========================================
  Files         496      510      +14     
  Lines       33655    36024    +2369     
==========================================
+ Hits        28981    30240    +1259     
- Misses       4674     5784    +1110     

☔ 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.