fix(bindings): correct poll_flush implementation

[lrstewart]

Resolved issues:

resolves #4803
If the issue doesn't make sense, check out the test I added to s2n_send_test.

Description of changes:

I fixed flush by allowing the bindings to call s2n_flush directly.

Callouts

Alternatively, we could "fix" s2n_send so that it can be used to flush. The problem is that the s2n_send behavior is deliberate and intended as a safety mechanism to detect an application calling s2n_send wrong. If we remove that safety mechanism, we make it slightly more likely that applications misuse s2n_send. I'm not sure just how much benefit customers get from this particular safety mechanism, but I'm also not sure how much benefit customers really get from flush :)

Another possible alternative would be to just make poll_flush a no-op. It's not super useful as-is since it can only clear the send buffer and not actually retry a full send. I don't know of any current valid use cases for poll_flush. But a no-op method seems stranger than a not particularly useful method :/ Maybe we can instead remove poll_flush before we make the bindings 1.0.0?

Testing:

The bug is kind of hard to imagine, so I wrote up a test to demonstrate why s2n_send isn't a replacement for s2n_flush. That should also prevent us from breaking s2n_flush in the same way, although that seems unlikely.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[lrstewart]

I vaguely remember that we didn't want to include "s2n.h" in this header, but I don't remember why and I couldn't find the discussion in the original PR.

However, I think this is necessary now because s2n_flush has s2n_blocked_status *blocked as an argument. s2n_blocked_status is currently an anonymous enum we typedef:

s2n-tls/api/s2n.h

Lines 2164 to 2172 in 9cd117f

	*/
	typedef enum {
	S2N_NOT_BLOCKED = 0,
	S2N_BLOCKED_ON_READ,
	S2N_BLOCKED_ON_WRITE,
	S2N_BLOCKED_ON_APPLICATION_INPUT,
	S2N_BLOCKED_ON_EARLY_DATA,
	} s2n_blocked_status;

The only way I could think to fix this without including "s2n.h" would be to switch s2n_blocked_status from an anonymous enum to a named enum, like in this example. But I'm not a fan of having two different ways to refer to the same enum.