Add a tracker

1. Create crates/uify-<name>/ with Cargo.toml depending on uify-core and nalgebra.
2. Pick the geometric primitive G and its covariance type C.
3. Implement uify_core::Tracker with Geometry = G, Covariance = C.
4. Add property tests for the associated manifold (if any) in crates/uify-core/tests/.
5. Add a synthetic GT scene for the new tracker under tests/synthetic/.
6. Register the crate in the workspace Cargo.toml and (if the CLAP plugin should expose it) in uify-clap-plugin.
7. Document at docs/reference/trackers/<name>.mdx.

Add a camera backend

1. Implement uify_runtime::camera::Backend in a new module uify-runtime/src/camera/<platform>.rs, feature-gated by #[cfg(target_os = "...")].
2. Deliver frames zero-copy where possible.
3. Add platform-specific system dependencies to nix/uify.nix.
4. Document at docs/development/platform.mdx.

Add an inference backend

1. Add an execution provider to uify-runtime::inference behind a Cargo feature.
2. Expose per-EP tolerances in the GT suite; do not compare numeric outputs across backends with ==.
3. Document at docs/reference/runtime/inference.mdx.

Add a transport

1. Create crates/uify-transport-<name>/.
2. Implement Sink<G, C> for the relevant Gs.
3. Document the wire schema at docs/reference/transports/<name>.mdx.

Add a language binding

1. Use UniFFI for Swift / Kotlin, pyo3 for Python, wasm-bindgen for web, napi-rs for Node.
2. Regenerate the binding artifact in CI and attach to releases.
3. Document at docs/reference/ffi/<language>.mdx.