Register Qwen 3.5 9B with the validated dense handler

[FurtherAI]

Summary

• Register Qwen/Qwen3.5-9B with ART's existing validated Qwen3.5 dense handler.
• Add a focused assertion for the model-to-handler mapping and validated native-vLLM LoRA status.

Qwen3.5-9B uses the same dense hybrid architecture already implemented by the
Qwen3.5 handler, so this does not change handler, runtime, LoRA, MTP, or
performance behavior.

The same registry entry was previously proposed in #806, but that stacked PR
targeted codex/gemma4-moe-lora-layout-fix after its route into main had
already landed. Its merge commit therefore never reached main.

Validation

• Commit gates: Ruff, Ruff format, ty, and lock consistency passed.
• Focused provider-support assertion: 1 passed.
• The full model-support workflow was not rerun because this only exposes an
  additional checkpoint name through the already-validated Qwen3.5 dense
  handler.