[TRANSFORMATIONS] Remove postponed_constant w/a in FuseMOEExperts

[mryzhov]

Details:

Fixes https://jira.devtools.intel.com/browse/CVS-182143 and https://jira.devtools.intel.com/browse/CVS-181151 should allow converting models with MoE experts without OOM errors. Therefore, we can remove setting the "postponed_constant" attribute on the v0::Concat node in the test model construction, aligning the test with the updated transformation logic.

Tickets:

• CVS-187700

AI Assistance:

• AI assistance used: no / yes
• If yes, summarize how AI was used and what human validation was performed (build/tests/manual checks).

[Copilot]

Pull request overview

This PR updates the FuseMOEExperts transformation and its unit test to remove the previous postponed_constant workaround, aligning the expected graph construction with the updated transformation behavior and aiming to prevent OOM during MoE expert weight fusion.

Changes:

• Removed the postponed_constant + disable_constant_folding workaround from FuseMOEExperts fused-weight construction.
• Updated the fuse_moe_test.cpp reference model to build fused expert weights via per-expert Unsqueeze + Concat, without setting postponed_constant.
• Minor formatting-only adjustments in extract_subgraph.cpp.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
src/common/transformations/tests/common_optimizations/fuse_moe_test.cpp	Updates reference model construction to match new MoE fused-weight graph (no postponed_constant, uses Unsqueeze + Concat).
src/common/transformations/src/transformations/utils/extract_subgraph.cpp	Formatting-only line wrapping changes.
src/common/transformations/src/transformations/common_optimizations/fuse_moe_experts.cpp	Removes postponed-constant workaround logic from MoE expert weight fusion.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.