There were two cases when the new async I/O worked incorrectly.
The problems only revealed themselves under specific
concurrency circumstances.

In one case, the client would get a mix of pieces of different
packets (responses).

The reason for this was that, when selecting which output buffer
to write to the client, the flushing algorithm used to select
whichever of two buffers, associated with a session, had data to
flush.
This worked OK as long as the entire contents of the buffer could
be written in one writev() call. But when writev() would only be
able to write a part of the buffer, the next writev() could
already be done with another buffer, thus producing a mix of
*partial* writes from both buffers.

Between two buffers used for a session, one is always newer than
the other (contains responses to requests which came in earlier).
So basically the bug was that we would select a newer
buffer for flushing, while there were still unhandled requests
in the older buffer.

One approach to a fix would be to, once started, keep writing
the newer buffer until there is nothing to write in it.

The actual fix chooses a simpler approach: to never begin
with a newer buffer, until the older one is completely written.
Thus, there is never reordering of output between two
output buffers, only within a single buffer.

The second bug was caused by wrong calculation of write offset
for iproto header. When iproto_header would fall on the
border of two iov vectors, obuf_book would rightfully
discard the tail of the first iov vector, and "position"
the header at the beginning of the second vector.
However, the savepoint used to memcpy() the output would
still point at the first vector, and the header would be
(partially) written at the end of the first vector,
not at the beginning of of the second.
This led to a) write beyond allocated memory b) loss
of a piece of a packet (effect seen by the client).

The fix is to correctly record the write position when
making a booking. obuf_book signature is changed
respectively.

No test cases since the bug is only spottable under certain
concurrency circumstances (a single CPU box, client/server I/O).
@todo: add unit tests.

Since connect/disconnect events in iproto run outside
fiber context, queue execution of on_connect/on_disconnect
triggers through the request queue. Generalize the request
queue for that purpose.

Implement a framework for on_connect/on_disconnect triggers
and Lua bindings for it.

Add tests and docs.

Move Box API declarations to box.h.
Make box.h a public header.
Add comments to box functions.
Rename fuctions accordingly.
Make includes for connector/c more local.

are not configured in the sequential order".

Fixing bug - absence of termcap and curses headers

Move the  block with @finally into an own function.

Implements
https://blueprints.launchpad.net/tarantool/+spec/fiber-specialization

After this patch, all requests even within the same session
are handled asynchronously, and responses are sent to the client as
soon as they become ready, regardless of the request position
in the input buffer.

This also changes  the internal implementation of iproto
from a "one fiber per connection" pattern to "fiber work crew"
pattern. This approach significantly reduces the amount of
context switches between fibers.

There is a bug in gcc Objective C support and nested exception,
with which an exception passes by a @catch clause if
there are nested @finally clauses. This patch doesn't work
around this bug, so if the server is compiled with gcc and
the client closes its side of the socket without fully reading
the input, the server terminates. The server needs to be compiled
with clang for now to work around this bug.

Do not modify output buffer between yield points:
in net-split, multiple fibers work on the same
session, and append output to the same output
buffer.