main tip does not build: the DSA schedule test predates W9's signature, and glm5_next_weights trips MSVC -Werror

[lu-zero]

Observed

Against main da2291339 (first executed on PR #2370's merge ref; main's own run had these jobs cancelled before executing, so the break was invisible there):

• build-test-cpu and both sanitize-cpu jobs fail compiling
  tests/vllm/models/test_glm_moe_dsa_schedule.cpp:304: cannot convert 'vllm::mla::MlaSharedSelection*' to 'vt::Tensor*'.
• Both windows-msvc-* jobs fail at
  src/vllm/model_executor/models/glm5_next_weights.cpp(223,31): error C2220 (warning treated as error).

Attribution

• The schedule test last changed in ee5c86031 (MODEL-TEXT-GLM-MOE-DSA W4); 11f34effb (the same row's W9, "pass W9's shared selection to the parameter it names") changed the signature without carrying the test. Owner: MODEL-TEXT-GLM-MOE-DSA.
• glm5_next_weights.cpp(223) last changed in a36add6a8 (MODEL-MM-GLM53-FLASH fix(MODEL-MM-GLM53-FLASH): read attention.key_length the way llama.cpp writes it, and DERIVE the KDA head count #2278). The in-flight row/BUILD-CPU-WERROR-* branches guard MXFP4 tests under VT_MARLIN_NVFP4 and do not touch this file. Owner: MODEL-MM-GLM53-FLASH.

Both facets block green CI for every open PR until repaired; neither is caused by any open PR's changes.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:zai-glm-5.3-flash [maki]

[lu-zero]

Both facets are on main, neither through the closed #2399: the schedule test's twelve-argument call landed as 08fa2f5 (#2395), and the KdaHeadCount rename (v_group/v_inner) landed byte-identical inside 5f23002 (ENGINE-HYBRID-PLACEMENT). Verified against the branch the fix was authored on: git diff origin/main row/BUILD-MAIN-TIP-TESTFIX -- src/vllm/model_executor/models/glm5_next_weights.cpp is empty. Closing.