feat(library): make LibraryLoader/DocumentLoader Send on native targets

[bkfunk]

Summary

LibraryLoader and DocumentLoader (in ldraw/src/library.rs) are declared with #[async_trait(?Send)]. That ?Send propagates through both native resolvers (LocalLoader, HttpLoader) and through resolve_dependencies_multipart.

On wasm32 this is necessary — wasm-bindgen futures are deliberately !Send because the web platform is single-threaded. But on native, the underlying work is already Send:

• LocalLoader uses tokio::fs::File / BufReader / try_exists — all Send.
• HttpLoader uses reqwest::Client futures — also Send.
• The shared PartCache lives in Arc<RwLock<...>> (already Send + Sync).

The ?Send was the only thing forcing native consumers running on a multi-threaded tokio runtime (e.g. Tauri's default) into ceremonies like:

tokio::task::spawn_blocking(move || {
    let rt = tokio::runtime::Builder::new_current_thread().enable_all().build()?;
    rt.block_on(async move { loader.load(&part_id).await })
}).await

A fresh runtime is constructed on every call. That's fine for one-shot loads but adds up for thumbnail-heavy UIs that batch dozens of part loads.

Change

Switch the four #[async_trait(?Send)] attributes (the two traits in library.rs and the four impls across resolvers/local.rs and resolvers/http.rs) to the standard cfg-gated pattern:

#[cfg_attr(not(target_arch = "wasm32"), async_trait)]
#[cfg_attr(target_arch = "wasm32", async_trait(?Send))]

On native, futures returned by LibraryLoader::load_ref, LibraryLoader::load_colors, and DocumentLoader::load_document are now Send. On wasm32, the existing ?Send behavior is preserved.

Compile-time Send regression guard

Adds dead helper functions in each resolver whose bodies call assert_send::<F>(_) on the futures returned by load_ref and load_colors. These never run, but the compiler type-checks their bodies, so any future regression that makes the futures !Send on native will fail compilation.

Test plan

• cargo check --workspace passes on native
• cargo check -p ldraw --target wasm32-unknown-unknown passes
• cargo test -p ldraw --lib passes

[Copilot]

Pull request overview

This PR removes the ?Send constraint on LibraryLoader and DocumentLoader traits (and their LocalLoader/HttpLoader impls) on native targets, while preserving ?Send on wasm32 where it is still required. This lets callers running on a multi-threaded tokio runtime (e.g. Tauri) .await these futures directly across threads instead of constructing a per-call current-thread runtime inside spawn_blocking.

Changes:

• Switch four #[async_trait(?Send)] attributes to a cfg_attr pair that uses the plain #[async_trait] (Send) on non-wasm32 and #[async_trait(?Send)] on wasm32, applied to both traits in library.rs and the four impls in resolvers/local.rs and resolvers/http.rs.
• Add per-resolver dead _assert_*_futures_are_send helpers gated to non-wasm targets that pass the futures returned by load_ref and load_colors to a generic assert_send::<F: Send> to provide a compile-time regression guard.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
ldraw/src/library.rs	Replaces ?Send on the DocumentLoader and LibraryLoader trait declarations with target-gated async_trait attributes.
ldraw/src/resolvers/local.rs	Same gating on the two LocalLoader impls; adds a non-wasm _assert_localloader_futures_are_send compile-time check.
ldraw/src/resolvers/http.rs	Same gating on the two HttpLoader impls; adds a parallel non-wasm _assert_httploader_futures_are_send compile-time check.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.