FEAT: Enable Local LLM Inference on Android

[iberi22]

Context

The current Android build uses a "Logic Engine" (mock) because ort (ONNX Runtime) failed to compile for aarch64-linux-android due to missing precompiled binaries/pkg-config issues.

Task

Enable real local inference on Android.

Options

1. Fix ORT: Configure ort to download/compile correctly for Android NDK.
2. Switch to Candle: Use Hugging Face's candle crate which is pure Rust and easier to compile for Android.
3. Switch to Llama.cpp: Use llama.cpp bindings.

Requirements

• Must run Phi-3-mini or Gemma-2b (quantized).
• Must not crash the app on load.
• Must stream tokens to the UI.

Reference

• crates/synapse-cognition/src/lib.rs
• apps/desktop/src-tauri/Cargo.toml

[google-labs-jules]

Jules is on it. When finished, you will see another comment and be able to review a PR.

[google-labs-jules]

Ready for a review! A PR has been created.