* fix(retry): also retry on non search methods when DNS failure

    Before this commit, methods with no fallback support (basically
every
    method) would fail at retry if the DNS error occured before the API
    client timeout.

    The behavior is now:
    - when a method with a fallback errors because of DNS failure we
      will
    switch to JSONP right away
    - when a method with no fallback errors because of DNS failure we
      will
    still retry

    We could use the same mechanism for both (= always try all hosts
    before JSONP) but I am not confident doing this change within a
patch
    or minor version. We have too litle data on how blocked XHRS are
    triggered (async, sync?).

    fixes #250

    * fix(request strategy): comply with retry spec, no early JSONP
    * switch

      This commit brings more conformance with request strategy
      specification by only raising timeout when there's a timeout.

      It also stop trying to switch to JSONP asap and always try all
hosts
      using XHRS before.