Add parse event init, node and done.

[mzimbres]

This PR implements parse events and is a prerequisite to the generic_flat_response from @D0zee.

[anarthal]

I don't think this is the right interface for any_adapter, because

• There's now a discrepancy between what adapters support (3 functions with different parameters) and what any_adapter is (a single function with an enum)
• Users can't create any_adapter objects, since the only practical way is using make_any_adapter, which is in detail.

Instead of this, I'd make the std::function and the enum an implementation detail, like this:

class any_adapter {
   enum class parse_event
   {
      /// Called before the parser starts processing data
      init,
      /// Called for each and every node of RESP3 data
      node,
      /// Called when done processing a complete RESP3 message
      done
   };

   using fn_type = std::function<void(parse_event, resp3::node_view const*, system::error_code*)>;
   fn_type impl_;

   template <class Response>
   static fn_type create_impl(Response& r)
   {
      using namespace boost::redis::adapter;

      return [adapter = boost_redis_adapt(
                 r)](parse_event ev, resp3::node_view const* nd, system::error_code* ec) mutable {
         switch (ev) {
            case parse_event::init: adapter.on_init(); break;
            case parse_event::node: adapter.on_node(*nd, *ec); break;
            case parse_event::done: adapter.on_done(); break;
         }
      };
   }

public:
   template <class T, class = std::enable_if_t<!std::is_same_v<T, any_adapter>>>
   explicit any_adapter(T& resp)
   : impl_(create_impl(resp))
   { }

   void on_init() { impl_(parse_event::init, nullptr, nullptr); };

   void on_done() { impl_(parse_event::done, nullptr, nullptr); };

   void on_node(resp3::node_view const& nd, system::error_code& ec)
   {
      impl_(parse_event::node, &nd, &ec);
   };
};

[anarthal]

I don't think this belongs here. You already maintain this information in multiplexer::elem::remaining_responses_. You can use this variable and go.

[anarthal]

I think that these should work as they are (without the make_).

[mzimbres]

@anarthal I think all concerns were addressed, please have a look at it again.

[anarthal]

Making this an Adapter (vs a Response&) is a breaking API change. To be honest, it makes sense for any_adapter to be constructible from an Adapter rather than a Response&, though. It is also true that all uses I've found end up calling boost_redis_adapt, so this feels like making syntax a little bit more convoluted. Is there any chance to make any_adapter(response) keep working?

[mzimbres]

I thought this was a small enough breaking change that wouldn't hurt anyone.

[anarthal]

It will break at least the user who originally requested any_adapter

[anarthal]

@mzimbres see my comments. I still have concerns about any_adapter's current interface.

[mzimbres]

@anarthal The response constructor in any_adapter has been reestablished. Please have a look.