Add Qwen3-VL tool calling support

[qgallouedec]

• Add response schema for Qwen3-VL (done by Claude)
• Add qwen3_vl.jinja template for identity matching in add_response_schema
• Add Qwen3-VL to TestAddResponseSchema and TestParseResponse test parametrizations
• Add Qwen3-VL to supported models in agent training docs

Part of #5460

Warning

Requires/contains #5459

---

Note

Low Risk
Low risk: adds a new Qwen3-VL chat template and wires it into existing response-schema selection, plus test/doc updates; minimal impact outside Qwen3/Qwen3-VL parsing paths.

Overview
Adds Qwen3-VL tool-calling support by introducing a new qwen3_vl.jinja chat template (including multimodal image/video placeholders) and treating it as Qwen3-equivalent for add_response_schema so parse_response can extract tool calls.

Updates unit tests to cover Qwen3-VL parsing/schema behavior and refreshes Qwen3 test fixtures, and documents Qwen3-VL as a supported model for GRPO agent training.

^{Reviewed by Cursor Bugbot for commit 535544b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[qgallouedec]

same template, but it's more natural to use the CausalLM instead of the SequenceClassification

[qgallouedec]

I can't remember why we used Qwen3MoeForSequenceClassification in the first place

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 91e940e. Configure here.}