GH-16719: Target encoding: Add feature interaction with known domain test is sometimes failing

[tomasfryda]

#16719

[Copilot]

Pull request overview

This PR addresses a failing test in the target encoding feature interaction functionality. The fix modifies the sparse categorical handling in NewChunk to preserve categorical type information for zero values when transitioning from sparse to dense representation, and adds a regression test to prevent the issue from recurring.

Key changes:

• Modified addCategorical() method to handle the special case where sparse zeros need to be marked as categorical before cancel_sparse() is called
• Added logic to preserve categorical information for zero values that are surrounded by other categorical values during sparse-to-dense conversion
• Added a regression test with 96 categorical values to reproduce and prevent the original multinode test failure

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 11 comments.

File	Description
h2o-core/src/main/java/water/fvec/NewChunk.java	Enhanced addCategorical() method to preserve categorical type information for sparse zero values during sparse-to-dense conversion
h2o-core/src/test/java/water/fvec/NewChunkTest.java	Added regression test to verify sparse categorical handling with specific data pattern that triggered the original failure

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[valenad1]

Thanks, nice, I am a bit nervous about changes in the core :) Could we rerun tests on latest rel-*?

[tomasfryda]

Sure @valenad1 ! Rerunning again. I hope it won't get aborted again.