Add memcpy binding

[szymon-zadworny]

This PR adds:

• Support for device-allocated buffers
• Generic Queue::copy binding
• An updated kernel launch example which showcases all current abstractions.

[Copilot]

Pull request overview

This PR expands the oneapi-rs USM buffer abstractions by adding device-allocated buffers and a generic Queue::copy (memcpy) binding, and updates the kernel launch example to demonstrate device→host transfers using the async event model.

Changes:

• Add a DeviceAllocator and a HostAccessible marker trait to differentiate host-accessible vs device-only USM buffers.
• Add Queue::alloc_device* and Queue::{copy, copy_with_deps} backed by a new SYCL queue::memcpy FFI binding.
• Update examples/kernel_launch.rs to allocate device buffers, launch a kernel, then copy results back to host and print.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
oneapi-rs/src/usm.rs	Introduces host-accessibility marker trait and exposes a public device allocator kind.
oneapi-rs/src/queue.rs	Adds device allocation helpers and new copy/memcpy APIs on Queue.
oneapi-rs/src/buffer.rs	Restricts slice deref to host-accessible allocators and adds buffer length accessor.
oneapi-rs/examples/kernel_launch.rs	Updates example to use async/await and device→host copy.
oneapi-rs-sys/src/queue.cpp	Adds C++ shim for queue->memcpy(...) with dependency events.
oneapi-rs-sys/src/queue-sys.rs	Exposes memcpy over cxx bridge to Rust.
oneapi-rs-sys/include/queue.hpp	Declares the new memcpy shim in the sys header.

Comments suppressed due to low confidence (1)

oneapi-rs/src/queue.rs:201

• Same as copy: please constrain T to Pod for this raw memcpy-based API so the safe surface area stays bitwise-sound.

    pub fn copy_with_deps<T, A1, A2>(
        &mut self,
        src: &Buffer<T, A1>,
        dst: &mut Buffer<T, A2>,
        dep_events: &[&Event],
    ) -> Event
    where
        A1: UsmAlloc,
        A2: UsmAlloc,
    {

---

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

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (2)

oneapi-rs/src/queue.rs:214

• size_of::<T>() is not in scope in this module, which will cause a compile error. Use a qualified path (or import it explicitly).

        let amount = min(src.get_len(), dst.get_len());
        let num_bytes = amount * size_of::<T>();
        unsafe {

oneapi-rs/src/usm.rs:34

• HostAccessible is declared as an unsafe trait, but it doesn't document the safety contract that implementers must uphold. For unsafe marker traits, add a # Safety section that spells out the required invariant (e.g., that allocations are directly readable/writable from the host for the lifetime of the allocation).

/// A marker trait for host-accessible USM allocators.
pub unsafe trait HostAccessible {}