qwen3-vl enablement

[libinta]

1. Fix the crash issue below as inputs_embeds. is expected to be 2D but the current logic returns inputs_embeds.shape=torch.Size([1, 1024, 5120]) when bs=1. bs>1 is not enabled yet for the flow

"/root/litang/github/qwen3/vllm/vllm/model_executor/models/qwen3_vl.py", line 1563, in _compute_deepstack_embeds
(EngineCore_DP0 pid=202) ERROR 12-08 15:59:28 [v1/engine/core.py:845] deepstack_input_embeds = deepstack_input_embeds.view(
(EngineCore_DP0 pid=202) ERROR 12-08 15:59:28 [v1/engine/core.py:845] RuntimeError: shape '[1, 3, 5120]' is invalid for input of size 3072
2. Enable multi-modal bucket warmup for qwen3-vl

[github-actions]

🚧 CI Blocked

The main CI workflow was not started for the following reason:

Your branch is behind the base branch. Please merge or rebase to get the latest changes.