ONNX and CoreML compilation pipeline

[bs258q]

Summary
Adds a model export pipeline for Needle that supports onnx, coreml, and tflite formats, with CLI integration, documentation, and a validation test.

Why
Enables mobile/edge deployment capability for Needle
Provides Apple CoreML support for iOS/macOS/ANE acceleration
Lays groundwork for full JAX→ONNX→CoreML/TFLite export once the complete toolchain is available

Notes
CoreML support is included and exposed through the CLI
Current implementation is intentionally simplified/placeholder-based; full production fidelity will require the full jax2tf / tf2onnx conversion toolchain

Testing
Verified python test_export.py under .venv
Confirmed export path and file creation for ONNX
Added documentation coverage for CoreML export usage
Raptor mini (Preview) • 1x

[shreyaskarnik]

Hi @bs258q sharing in case it's useful: I've been working on a similar export from a different angle and ended up with a browser-ready ONNX pipeline for Needle (encoder + decoder w/ KV cache, Flax↔PyTorch↔ONNX parity checks, SentencePiece port).

• Playground: https://huggingface.co/spaces/shreyask/needle-playground
• ONNX weights: https://huggingface.co/onnx-community/needle-onnx
• Porting guide + scripts: https://huggingface.co/onnx-community/needle-onnx/blob/main/PORTING.md and https://huggingface.co/onnx-community/needle-onnx/blob/main/export_onnx.py

It's out-of-tree (lives in my own repo, not upstreamed), but if any of it's worth folding into this PR, feel free to lift whatever's useful happy to answer questions.