Move TorchVision App Demo to `pytorch_android_lite`

🚀 The feature

Currently we depend our Android Demo app depends on pytorch_android. It's recommended to switch to pytorch_android_lite.

Transitioning to the lite version requires us to update the following things:

• Change the exported file from frcnn_mnetv3.pt to frcnn_mnetv3.ptl
• Export the fasterrcnn_mobilenet_v3_large_320_fpn model using _save_for_lite_interpreter()
• Update the dependencies to org.pytorch:pytorch_android_lite and org.pytorch:pytorch_android_torchvision_lite
• Replace Module.load() with LiteModuleLoader.load() on the code-base

Motivation, pitch

The recommended dependency for mobile is pytorch_android_lite.

Alternatives

No response

Additional context

There are a few bugs that block this at the moment:

• The release pipeline for pytorch_android is buggy. The -SNAPSHOT version contains only the _lite binaries and there is no snapshots for pytorch_android_lite.
• There seems to be a bug on the serialization mechanism that exports the model. See https://github.com/pytorch/pytorch/issues/68536

cc @IvanKobzarev

Start by locating the Android Demo, its model-export code, dependency declarations, and uses of Module.load(). Check the release-pipeline and serialization blockers, including issue #68536, before making the transition. Done means the demo exports the .ptl model, uses the lite dependencies and loader, and builds successfully.