[BUG]Incompatible locally imported models remain inaccessible and undeletable

[weiyuanyue]

Currently, when users import a local ONNX model into AI Dev Gallery, the compatibility check is only performed after the model has been successfully imported. For models requiring DML/GPU acceleration on x64 devices, the compatibility logic enforces strict requirements:

1. A dedicated GPU must be present
2. Available VRAM must exceed model size by at least 2GB (for compatibility)

If these requirements are not met, the model is marked as [NotCompatible]. However, this creates a critical UX issue:

Current Behavior

1. User successfully imports a local model file → ✅ Import succeeds
2. System evaluates compatibility post-import → ❌ Model marked as NotCompatible
3. Model becomes invisible in the Model Management UI
4. User cannot use the model (expected)
5. User cannot delete the model (problematic)

One of the following approaches should be implemented:

Option 1 : Pre-import Validation

• Perform compatibility checks before allowing model import
• Display clear warning messages if hardware requirements are not met
• Allow users to proceed with import only after acknowledging the incompatibility

Option 2: Post-import Management

• Allow incompatible models to remain visible in Model Management UI
• Display compatibility status clearly (e.g., with warning badges)
• Enable deletion of incompatible models

Option 3: Hybrid Approach

• Pre-import warning with user confirmation
• Post-import visibility with clear status indicators

Impact

• User Confusion: Users don't understand why their successfully imported model is nowhere to be found
• Storage Waste: Incompatible models consume disk space but cannot be managed
• Poor UX: No feedback loop or recovery mechanism for users who inadvertently import incompatible models