Use MobileSAM on Android, enable NNAPI and unload unused AI models fast

[waxm134]

Description

• AI model performance tweaks and bug fixes for Android

Type of Change

• Bug fix
• New feature
• Breaking change
• Performance improvement
• Code refactoring
• Documentation update
• UI/UX improvement
• Build/CI or Dependency update

Changes Made

• SAM2 model is heavy for android devices, switch to MobileSAM on Android to increase subject mask speed and lower RAM usage
• Enable NNAPI to leverage any NPU provided by the SoC
• unload unused AI models quickly to avoid OOM fails due to excessive RAM usage
• fixed inpainting on Android

Screenshots/Videos

Testing

• I have tested these changes locally and confirmed that they work as expected without issues

Test Configuration:

• OS: Ubuntu 24.04, Android 16
• Hardware: Intel i5, Pixel 9 Pro, Xiaomi Pad 7

Checklist

• My code follows the project's code style
• I haven't added unnecessary AI-generated code comments
• My changes generate no new warnings or errors

Additional Notes

• I did not notice any worsened performance of the Subject Mask tool with MobileSAM compared to SAM
• heavily reduces RAM usage on Android (Subject masks always made my Xiaomi Pad 7 crash due to OOM, now it runs fine)
• MobileSAM is currently embedded from this repo: https://huggingface.co/Acly/MobileSAM, should be added to the RapidRAW AI model repo (don't know what the procedure is there)
• MobileSAM git: https://github.com/ChaoningZhang/MobileSAM/tree/master
• while testing this I discovered an existing bug with the inpainting on android (inpainted area was always just black)
• inpainting can be fixed by using LAMA_Dilated instead of LAMA, as Android seems to have issues with the fp_16 operations of the LAMA model
• LAMA_Dilated cannot use NNAPI so it must be initialized without it, works on Pixel 9 Pro as well even if its not a Snapdragon processor so I guess it should work on non-Qualcomm processors in general as well
• LAMA_Dilated is currently embedded from this source: https://huggingface.co/qualcomm/LaMa-Dilated/tree/ab898502c9bd764a50eb2719a309694b43eae658
• LAMA_Dilated git: https://github.com/qualcomm/ai-hub-models/tree/main/src/qai_hub_models/models/lama_dilated

AI Disclaimer:

Please state the involvement of AI in this PR:

• This PR is entirely AI-generated
• This PR is AI-generated but guided by a human
• This PR was handwritten with AI assistance (spell check, logic suggestions, error resolving)
• This PR contains only blood, sweat, and coffee (AI-free)

[waxm134]

Moved this to a new branch on my fork, closing in favor of #1090