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.

[iberi22]

Actualizacion 2026-03-03: PR #703 fue cerrado por no-mergeable (dirty/conflictos). Este issue sigue abierto para retomar en rama nueva rebased desde main.

[iberi22]

Actualización de dependencia: este issue queda bloqueado por la decisión/avance en runtime de inferencia (#680 ORT vs #676 Candle+RWKV) y por integración móvil del epic #176.

[github-actions]

✅ Auto-assigned to @iberi22 (label: in-progress)

[iberi22]

Removing blocked label - today's Candle migration PR (#753) addressed Android inference path. If ort/llama.cpp still needed, please confirm or re-add blocked label.

[iberi22]

⚠️ Stale Issue - No updates for 28+ days.

This issue is marked in-progress with jules but has had no activity since 2026-03-03.

Status check: Is this still being worked on? If Jules is actively handling it, please update. Otherwise, should we close this as stale or move to a backlog?