[WIP] Use miniz_oxide directly in rust backend

[oyvindln]

This avoids the C API and it's unsafeness entirely.

Can probably further simplify a bit by altering the Stream struct and direction trait StreamWrapper doesn't have to be an enum, neither are public.

Needs some thorough testing.

fixes #181

[alexcrichton]

Thanks for this! One thing I'd like to do here is to reduce the number of #[cfg] necessary. Would it be possible to push more of the #[cfg] into the src/ffi.rs module itself? That way the backend-specific implementations could all be located there and ideally src/mem.rs is free from #[cfg] (largely at least)

[oyvindln]

I'll look into it.

[oyvindln]

Right, pushed all of the back-end specific stuff into ffi.rs. Only ones that remain are the zlib-specific public functions.

Is there some specific reason the zlib mz_stream is boxed, while the miniz one is not? Otherwise they could share that code.

[alexcrichton]

That's actually probably a bug, want to go ahead and box both of them up? I think there's also an error on nightly to handle?

[oyvindln]

I'll box both then.

The nightly error may actually be something that's in the original code, the lint was added recently, and the original code uses mem::zeroed() here too. I suspect it's the extern fn fields in z_stream that the compiler does not want to assume zero is a valid value for. In the miniz variant, those fields are wrapped in option.

[alexcrichton]

Hm yeah so I think I messed that up when I wrote libz-sys.

If you wouldn't mind adding #[allow(invalid_value)] for now I can try to clean this up after this lands.

[alexcrichton]

👍