fix(mm): detect FLUX.2 Klein 9B Base variant via filename heuristic

[Pfannkuchensack]

Summary

FLUX.2 Klein 9B Base (undistilled) and Klein 9B (distilled) have identical model architectures — same block counts, same tensor shapes, same keys. They cannot be distinguished from the state dict alone.

This PR adds a filename heuristic to detect the Base variant: if the model name contains "base" (e.g. flux-2-klein-base-9b-fp8.safetensors), it is classified as klein_9b_base instead of klein_9b.

Also fixes the diffusers format detection, which incorrectly relied on the guidance_embeds config field to distinguish the variants. Both variants have guidance_embeds=False.

QA Instructions

1. Import a FLUX.2 Klein 9B Base checkpoint with "base" in the filename (e.g. flux-2-klein-base-9b-fp8.safetensors)
2. Verify it is detected as klein_9b_base
3. Import a FLUX.2 Klein 9B (distilled) checkpoint without "base" in the filename
4. Verify it is detected as klein_9b
5. Verify Klein 4B models are unaffected

Merge Plan

No special considerations.

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)

[lstein]

Works as advertised.