Add `poll_for_reply` and `poll_for_reply_unchecked` functions

[Antikyth]

This PR adds poll_for_reply and poll_for_reply_unchecked functions - the async equivalents of the blocking wait_for_reply and wait_for_reply_unchecked. The poll API is based on the difference between wait_for_event and poll_for_event.

Where wait_for_reply and wait_for_reply_unchecked are the Rust bindings to xcb_wait_for_reply64, poll_for_reply and poll_for_reply_unchecked are the Rust bindings to xcb_poll_for_reply64.

Closes #244.

[Antikyth]

The changes to wait_for_reply and wait_for_reply_unchecked are just:

• Documentation changes
  • Fixed spelling mistake: "occured" -> "occurred".
  • Explained that they are the blocking versions, and linking to the non-blocking versions (the non-blocking versions do the same, linking to the blocking versions).
• Changed std::ptr::null_mut() to ptr::null_mut(), as std::ptr is imported and ptr::null_mut() is used in other places.
  • Doesn't affect the behavior of the code at all.

[sprocklem]

Hello @Antikyth, poll_for_reply_unchecked still seems to have the wrong behaviour. The documentation in the header is a little misleading here, but xcb_poll_for_reply* return 1 if XCB has received a corresponding reply or error, or 0 if no response of either type has been received. A return value of 1 with a null reply (where the current code panics) is perfectly valid: it is returned if an error was received by the client even though the error is not returned by the function.

This is more of a personal opinion thing, but it also seems to me that the Options are inverted: it seems like a request that is still waiting on a response should return Ok(None), while a request that has received an error (and, therefore, has not reply) should return Ok(Some(None)).

Edited to add: xcb_poll_for_reply* also returns 1 with both reply and error set to null if there is a connection error. You should insert a self.has_error()? after getting a return value of 1 in both functions, because the current code panics on connection error.

The return value is really a "would not block" boolean, and should be treated as such.

[rtbo]

@sprocklem is right, the return value of xcb_poll_for_reply indicates whether the request is finished. If request is finished we have to check if it is a reply or an error.

Therefore the return value of Connection::poll_for_reply should be Option<Result<C::Reply>> rather than Result<Option<Reply>>. And to be consistent with Connection::wait_for_reply_unchecked, Connection::poll_for_reply_unchecked should return Option<ConnResult<Option<C::Reply>>>.

To summarize

	checked	unchecked
blocking	wait_for_reply -> Result<C::Reply>	wait_for_reply_unchecked -> ConnResult<Option<C::Reply>>
non-blocking	poll_for_reply -> Option<Result<C::Reply>>	poll_for_reply_unchecked -> Option<ConnResult<Option<C::Reply>>>

[rtbo]

Closed in favor of #248. Thank you @Antikyth for bringing this up!