v1.6.0 - Qwen-Image-Edit support

Highlights

• Qwen-Image-Edit support (and other temporal / 5D-latent models like Wan), now a first-class edit model alongside FLUX 2 Klein. Generate, Refine, Xtra-Fine, Area Prompt, Smart Inpaint, and Smart Guided Inpaint all work. Fixes #14.

Fixes

• Qwen/Wan 5D latents — the VAE boundary is centralised so 5D [B, C, T, H, W] decode output is folded for previews, and the base latent is normalised to the dimensionality each model expects before sampling (the step stock KSampler does, which the node was bypassing).
• Smart Inpaint feathering on Qwen/Wan — no longer distorts. The feather now lives in the pixel composite + latent blend (the standard Qwen-edit recipe) rather than the denoise mask; 5D sampling uses a binary mask. FLUX/Klein behaviour is unchanged.
• Preview width — the preview/controls now fill the full node width instead of snapping to a narrow column on the left when you click into the node.

Docs

• README rebalanced: FLUX 2 Klein 9B and Qwen-Image-Edit are presented as co-equal first-class edit models.
• Added a bundled Qwen-Image-Edit 2511 example workflow (workflows/Qwen Edit 2511 example.json).