[CanReader]

What this changes

README examples now compile as written, the stale bits are gone, and the crate description is plainer: "A small deep learning library in Rust with optional CUDA kernels".

Why

I compiled every snippet in the README before the first publish. Three were wrong:

• the quick start used ? outside a function, so it is wrapped in main now
• the layer example called scoped without importing it (fastnn::nn::module::scoped)
• the op example used a fastnn_cuda_softplus kernel that does not exist, so it shows sqrt as it is in the code

Also drops the Hacktoberfest paragraph (the README ships to crates.io and will go stale), adds a note that 0.x can still change, mentions stubs.c in the layout, and removes the "CUDA 13 runtime" claim I could not back up.

How it was verified

• cargo test passes
• cargo clippy --all-targets -- -D warnings is clean
• Ran cuda_parity on a GPU
• Trained an example in --release
• No GPU here, CUDA side not runtime tested

Docs only, plus the description line in Cargo.toml. cargo package still builds.

Checklist

• Every new or changed backward rule has a gradcheck case
• Every new kernel has a cuda_parity case
• tensor/ops and autograd/ops still mirror each other
• Public items have /// docs