go bindings: should we try to use rustgo (no cgo)?

[cyphar]

cgo function calls are quite expensive (~20x) and so there might be an argument that we should try to do some Go assembly hacks to try to avoid using cgo. On the other hand, there are obvious maintenance issues with magic assembly code, and it's not clear that the (real) overhead of cgo is a practical issue.

Food for thought...

Some links:

• https://words.filippo.io/rustgo/ (note the 2024 update that you want to use syso).
• https://go.dev/wiki/GcToolchainTricks
• https://go.dev/doc/asm