Adds a sans-io fsm to the read operation.

[mzimbres]

@anarthal: Notes for the code review

• Adds a fsm for read operation following your work on exec_fsm.
• Log every event of the reader_fsm after it resumes, which improves log information content.
• The logger interface has to be extended with types that are actually in the detail namespace, so I make fsm actions a public type, this is needed to interpret log messages. I would like to do this to the exec_fsm too once a log object becomes available for async_exec, so I did not change your code yet.
• Cancellation hasn't been addresses yet for read operation. This will need more thought and a separate PR.

Example log for the subscriber example. Once all ops are refactored to use an fsm, the log will become very detailed and more helpful

(Boost.Redis) resolve results: 127.0.0.1:6379
(Boost.Redis) connected to 127.0.0.1:6379
(Boost.Redis) fsm action: (reader_action::type::setup_cancellation, 0, Success)
(Boost.Redis) fsm action: (reader_action::type::append_some, 0, Success)
(Boost.Redis) writer_op: 117 bytes written.
(Boost.Redis) fsm action: (reader_action::type::notify_push_receiver, 36, Success)
(Boost.Redis) fsm action: (reader_action::type::append_some, 0, Success)
(Boost.Redis) hello_op: Success
subscribe channel 1
^C
(Boost.Redis) ping_op (4): Operation canceled
(Boost.Redis) check_timeout_op (2): Operation canceled
(Boost.Redis) fsm action: (reader_action::type::cancel_run, 0, Success)
(Boost.Redis) fsm action: (reader_action::type::done, 0, Operation canceled)
(Boost.Redis) writer_op (3): connection is closed.

[anarthal]

This looks good in general. In the longer run, I'd like to:

• Avoid the cancel_run action completely in the reader and writer - I believe code would be cleaner if it was the parallel group that handled this.
• Dismantle async_append_some and merge it into the reader (async_compose has some overhead).

But this is not in scope for this PR.

[mzimbres]

I think I addressed all your points. Please have a look at it again. Are you about to get the log PR merged? I will wait and then rebase on top of your changes so that I can remove the usage of types in detail in the logger interface.

[anarthal]

Looks good. I've still some work to do, as CIs don't pass yet and there's a ton of tests to write. I think it's better if you merge/squash merge it and I'll move all action types back to detail as part of the logger PR.