spmc 0.2.3 has been yanked on 2019-08-15

Passing to 0.2 requires a small change in the tests.
Changing the following (as was done in master branch)
  &'a spmc::Sender<Reply> -> &'a Mutex<spmc::Sender<Reply>>

This fixes the cargo issue telling us that he cannot handle smpc:
  error: failed to select a version for the requirement `spmc = "^0.2"`

Signed-off-by: Gabriel Ganne <[email protected]>

Closes #1925

(cherry picked from commit bb365f6)

`bytes` is bumped due to the addition of `Bytes::advance` in 0.4.6. `h2`
0.1.13 introduced the `.is_io()` and `into_io()` methods.

Bumping `tokio-threadpool` to 0.1.16 removes a dependency on `rand` from
the crate which reduces the size and avoids minimum version dependency
issues in that dependency tree.

This takes the same strategy as golang, where the timeout value is
divided equally between the candidate socket addresses.

If happy eyeballs is enabled, the division takes place "below" the
IPv4/IPv6 partitioning.

Backported to 0.12 from master.

…utor (#2009)

add generic for the executor to server impl block

resolves #1988

The recently-added connect timeout does not cover resolving hostnames, which
could also stall on an OS-level timeout if there are issues reaching the DNS
server.

…aders

When a request arrived with multiple `Transfer-Encoding` headers, hyper
would check each if they ended with `chunked`. It should have only
checked if the *last* header ended with `chunked`.

See GHSA-6hfq-h8hq-87mf