Add WebGPU compute crate scaffold and wasm-pack integration

[seanwevans]

Motivation

• Prepare the codebase to host a browser-native WebGPU compute backend by adding a dedicated workspace crate and a shared GPU backend contract so CUDA and WebGPU implementations can interoperate.
• Provide a minimal, runnable WebGPU initialization + dispatch path so the React frontend can load a wasm artifact and exercise compute dispatch while the FP32 CUDA SGP4 kernel is ported to WGSL.
• Surface an explicit build flow for producing a WebAssembly package consumable by the web client via wasm-pack.

Description

• Add a new workspace member openastroviz-webgpu-compute with wgpu + wasm-bindgen scaffolding, including webgpu-compute/src/lib.rs, webgpu-compute/Cargo.toml, webgpu-compute/README.md, and webgpu-compute/shaders/sgp4_fp32.wgsl as the WGSL kernel placeholder.
• Introduce GpuBackend trait and GpuBackendError enum into core (core/src/lib.rs) to formalize the CUDA/WebGPU parity points and provide dispatch_sgp4_fp32_step semantics.
• Implement WebGpuBackend that requests an adapter/device/queue, compiles the WGSL shader into a compute pipeline, and exposes wasm-bindgen exports initWebGpuBackend and dispatchSgp4Fp32Step for browser-side invocation.
• Wire a wasm build step into the web package by adding the build:wasm script to web/package.json and documenting yarn build:wasm/wasm-pack usage in web/README.md and the crate README.

Testing

• Ran cargo fmt and formatting completed after installing rustfmt (succeeded).
• Ran cargo check which built the workspace including the new openastroviz-webgpu-compute crate (succeeded).
• Ran cargo test -p openastroviz-core and all core unit/regression tests passed (2 Vallado vector tests passed).

---

Codex Task