Add the multi wakeup socket (when available) to the pollset
        of the admin handle *when* transfers are alive. This markes
        the admin handle as "dirty" on socket events and lets it run.

        Without this, event based processing never sees the wakeup socket
        and never acts on it. I assume the wakeup socket never worked for
        event based processing.

        What is this good for?

        In a future PR, based on this one, we can trigger async resolver
        results to use the multi's wakeup socket and save the transfer
        based wakeup sockets. This also solves lifetime problems for such
        sockets when a transfer is destroyed early, as the multi wakeup
        system remains unchanged during the multi's lifetime.

        Give the dnscache implementation its own source file, as
        hostip.c became unwieldy.

        have Curl_dnscache_get() return a CURLcode for negative entries
        fix include for unit1305.c

        - remove `struct Curl_dns_entry *dns` entry
        - remove `done` BIT
        - make port an uint16_t
        - make ip_version an uint8_t
        - rename `Curl_async_is_resolved()` to `Curl_async_take_result()`,
          returning CURLE_AGAIN while waiting on the result
        - remove `Curl_once_resolved()`

        Turn the fixed easy handle member `async` into an allocated struct.
        We allocated the contained hostname already, so fold that into the
        struct itself and make the easy handle a little smaller.