Fix: Z-Image LoRA detection for Kohya and ComfyUI formats

[Pfannkuchensack]

Summary

Add support for Kohya-format Z-Image LoRAs (lora_unet__ prefix) by adding key detection and conversion to dot-notation module paths. Fix ComfyUI-format Z-Image LoRAs being misidentified as main models by ensuring LoRA-specific suffixes (including .alpha) are checked before Z-Image key matching in _has_z_image_keys().

Related Issues / Discussions

https://discord.com/channels/1020123559063990373/1149506274971631688/1487673057400197180

QA Instructions

Test with two Z-Image LoRA files in different formats:

Kohya format (keys like lora_unet__layers_0_attention_to_k.lora_down.weight):

• Should be auto-detected as LoRA / LyCORIS / Z-Image
• Keys should be converted to dot-notation (layers.0.attention.to_k) and applied correctly during inference

ComfyUI format (keys like diffusion_model.layers.0.attention.to_k.lora_down.weight):

• Should be auto-detected as LoRA / LyCORIS / Z-Image (previously misidentified as Main / Checkpoint / Z-Image)
• Should apply correctly during inference

Regression checks:

• Z-Image main model checkpoints should still be detected as Main / Checkpoint / Z-Image
• FLUX, SD1, SD2, SDXL LoRA detection should be unaffected

Merge Plan

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable)
• ❗Changes to a redux slice have a corresponding migration
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)