fix unsafe.Pointer use to be compatible with cgo rules

[calebdoxsey]

According to golang/go#14210 it's necessary to take the address of a byte slice in the same expression as the cgo function invocation.

IE do this:

C.f(unsafe.Pointer(&dst[0]))

And not this:

tmp := unsafe.Pointer(&dst[0])
C.f(tmp)

This PR will fix #24 without resorting to CBytes, and therefore avoiding a copy.

[ghost]

@confluentinc It looks like @calebdoxsey just signed our Contributor License Agreement. 👍

Always at your service,

clabot

[edenhill]

This is great stuff, will tend to this soon.
Thank you

[edenhill]

There is a difference between an empty Value/Key field (length 0) and a Null message (Value/Key is nil).
Empty messages are treated like any other message, just empty, while Null messages values are considered tombstones and null message keys are considered key-less.
That's the reason why the original code initialized both valp and keyp to nil (NULL in C) and I'm guessing this will break the cast-on-same-line requirement of Go.
One way around this might be to have four different code paths depending on the nullity of key and value, e.g.:

 if  key && val {
    produce(..., key, val, ..)
 else if key && !val {
   produce(..., key, C.NULL, ..)
 ..and so on

[edenhill]

Fixed in #83

Thank you for your PR, much appreciated!