Qwen3 VL support (text only)

[BBC-Esq]

Created a loader for Qwen3VL, which is very similar to the Qwen3 loader with some minor naming differences. Qwen3VL is somewhat superior to Qwen3, even though the names are similar. Even when using just for text, Qwen3-VL will beat a regular Qwen3 model...I've tested this, but at any rate here's what the HF repo gives as an example:

• Initially supported in Transformers 4.57

NOTE: I don't think it's going to be possible to create an AWQ support, but it's still worth supporting text only.

Category	Benchmark	Qwen3-VL 8B Instruct	Qwen3-8B Non-Thinking
Knowledge	MMLU	80.7	77.6
Knowledge	MMLU-Pro	71.6	63.4
Knowledge	MMLU-Redux	84.9	79.5
Knowledge	SuperGPQA	44.5	35.8
Reasoning	AIME25	45.9	20.9
Reasoning	HMMT25	32.5	11.8
Reasoning	LiveBench1125	62.0	53.5
Code	LCBv6 (25.02–25.05)	39.3	25.5
Instruction Following	SIFO	47.1	45.7
Instruction Following	SIFO-multiturn	38.3	37.7
Instruction Following	IFEval	83.7	83.0
Subjective Eval	WritingBench	83.1	71.4
Agent	BFCL-v3	66.3	60.2
Multilingual	MMLU-ProX	65.4	58.0
Multilingual	INCLUDE	67.0	62.5
Multilingual	PolyMATH	30.4	18.8

I apologize for the other PRs regarding examples...I can't figure out how to remove them from this PR because is coming from a branch of my repo but my "main" branch already accepted them so...NOT a github expert by any means. Just delete or what not, or better yet accept them!