[performance] Use buffers instead of channels

Use buffers instead of channels, maybe https://crates.io/crates/crossbeam could help or we can impl a channel like (to be easier to use) which has a fixed buffer where it keeps data and tx sends offsets and rx exposes slices based on those offsets, as we only need to access the buffer for reading in rx

No files, tests, or entry points are named. Start by locating the channel, tx, and rx usage in the Rust implementation, then compare the existing data flow with crossbeam's buffering model. Done would require an agreed buffer design and evidence that it improves performance without changing the transport behavior.

Written by the indexing model from the issue text.