ops: limit return of requants (Fix perf of some fp8 dynamic_vram workflows)

[rattus128]

This check was far too broad and the dtype is not a reliable indicator of wanting the requant (as QT returns the compute dtype as the dtype). So explictly plumb whether fp8mm wants the requant or not.

Example Test Conditions:

Windows, RTX5060, 32GB RAM, --fast dynamic_vram
WAN2.1 fp8 scaled 14B + Lora

Before:

Requested to load WAN21
0 models unloaded.
Model WAN21 prepared for dynamic VRAM loading. 13630MB Staged. 1053 patches attached.
100%|████████████████████████████████████████████████████████████████████████████████████| 4/4 [03:38<00:00, 54.52s/it]
Requested to load WanVAE
Model WanVAE prepared for dynamic VRAM loading. 242MB Staged. 0 patches attached.
Prompt executed in 250.19 seconds

After:

Requested to load WAN21
0 models unloaded.
Model WAN21 prepared for dynamic VRAM loading. 13630MB Staged. 1053 patches attached.
100%|████████████████████████████████████████████████████████████████████████████████████| 4/4 [03:16<00:00, 49.10s/it]
Requested to load WanVAE
Model WanVAE prepared for dynamic VRAM loading. 242MB Staged. 0 patches attached.
Prompt executed in 229.95 seconds