Better error handling

[LaurentMazare]

It would be great if errors could be used in favor of panics and also if bindgen_cuda::Error could implement the std::error::Error trait.

[Narsil]

What does it change for builds ? I started doing that but the error messages where not really great when I started doing so, since it swallowed the stack and I don't really want to pull anyhow to get the backtrace (which pretty much does just like a panic).

I don't really see the benefit of non panicking during builds, especially if the error is unrecoverable (like cuda cannot be found).

[LaurentMazare]

I guess it leaves the flexibility of panicking or not to the library's users rather than deciding what's the best for them (and if they only want to use this for builds or something else). The error message being explicit and different I'm not sure the backtrace would be worth it indeed so would just go with a very simple error type.

[Narsil]

Fair enough, bindgen also has a CLI anyway which I think can come in handy and we'd definitely need errors there too.