[Gemma 4] Refactor Gemma 4 weight loader and qkv proj

[kpham-sgl]

Motivation

Refactor QKV proj to avoid wasted KV proj for KV shared layers

Modifications

• KV-shared layers now load and call only a separate q_proj (ColumnParallelLinear) — no fused qkv_proj, no k_norm / v_norm.
• Hoisted the entire KV-sharing decision (including kv_shared_layer_index lookup) into Gemma4DecoderLayer.__init__; Gemma4Attention just receives the resolved index.

Accuracy Tests

Model	Reference (cookbook)	This run	Δ
gemma-4-E2B-it	0.307	0.296	−0.011
gemma-4-E4B-it	0.396	0.402	+0.006
gemma-4-31B-it	0.589	0.576	−0.013
gemma-4-26B-A4B-it	0.549	0.559	+0.010

Speed Tests and Profiling

[TODO] profile to make sure K_proj and V_proj does not appear for the smaller models

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[kpham-sgl]

Close in favor of #25461