feat(model): Add Config Option To Enable fp32 Precision For Anima

[kappacommit]

Summary

Adds an anima_precision: auto | bfloat16 | float32 field to invokeai.yaml so users can force Anima into fp32. Default auto preserves today's bf16-with-safe-fallback behavior bit-for-bit; float32 flips precision across the whole Anima pipeline — model loader, denoise invocation, and Qwen3 text-encoder LoRA patches.

Motivation

On Apple Silicon (MPS), bfloat16 is allocatable but slow, producing ~6x slowdowns vs ComfyUI on the same Anima checkpoint. ComfyUI users work around this with --force-fp32; this PR is the equivalent escape hatch.

Usage:

# in invokeai.yaml
anima_precision: float32

Why a dedicated field instead of reusing the existing precision?

Flux uses an existing field, precision to allow this. However, precision: auto (the default) resolves to
fp16 and Anima is bf16-native: fp16's narrower exponent range produces broken output. So a literal "make Anima respect global precision" would silently break Anima for every user on default config.

Extension path for other bf16-native models
Z-Image and Qwen Image currently hardcode bf16 the same way Anima did. Adding the same knob for either is a small mechanical change:

1. Add a literal alias and field to InvokeAIAppConfig
2. Replace TorchDevice.choose_bfloat16_safe_dtype(device) in that model's loader/invocation with TorchDevice.resolve_model_precision(get_config().z_image_precision, device).

No changes to resolve_model_precision itself.

Related Issues / Discussions

Discussed in Discord https://discord.com/channels/1020123559063990373/1149506274971631688/1500552718455148735

QA Instructions

Configure your invoke.yaml:

  anima_precision: float32

Restart invoke and generate an image with Anima. Confirm successful generation. Note it may be slower on lower VRAM machines.

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)

[kappacommit]

User confirmed the fix here: https://discord.com/channels/1020123559063990373/1149506274971631688/1500611525810065558