Follow-up to #5353: Move another condition within views::filter

[Priya2698]

No description provided.

[github-actions]

Review updated until commit fa81b5e

Description

• Fix stream condition logic in loop domain filtering

• Remove redundant stream check in transform expr loop

• Apply consistent stream parallelization condition

• Simplify allocation domain transformation logic

---

Changes walkthrough 📝

	Relevant files
Bug fix	finalize_multidevice_domains.cpp Refactor stream and allocation domain logic                            csrc/preseg_passes/finalize_multidevice_domains.cpp Move stream parallelization condition into loop domain filter Remove redundant stream check in transform expr processing Use consistent shouldParallelizeAllocationOnStream condition Simplify allocation-to-contiguity update logic +3/-6

---

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🧪 No relevant tests

⚡ Recommended focus areas for review Logic Change The filtering condition for stream dimensions has been moved from the loop iteration (post-split check) to the view definition (pre-split filter). This changes when shouldParallelizeAllocationOnStream(tv) is evaluated and may affect which expressions are processed. The impact on sharding correctness and allocation propagation should be validated, especially for cases where stream parallelization is disabled. tv->getLoopDomain() | std::views::filter([&tv](IterDomain* id) { return id->isDeviceDim() || (id->isStream() && shouldParallelizeAllocationOnStream(tv)); });

[Priya2698]

!test