Better debug options

Nobody has claimed this yet.

Working on #42 it was very hard to figure out what was happening if something didn't work as expected. For example: one of my theory on #40 is with awkward timing, a connection may be closed just when fin-ack [is sent](https://github.com/bittorrent/libutp/blob/master/utp_internal.cpp#L2181), which may mean `utp_call_on_state_change` may never be executed, causing utp-native to wait until timeout until closed.

The reason why I came this far is because I enabled `libutp`'s debug mode and added a _lot_ of print statements to the bindings, as well as adding an identifier to the connection so I could make sure in the log statements _what_ operation caused a problem.

Adding/Removing the log statements to find the cause of an error is stressful. Which is why I would like to propose to add debug support to the library itself so debugging can be easier.

- Add an incrementing number to the `utp` and `connection` instances that is passed-on to the bindings which allows to figure out _which_ connection is doing what in a command line.
- Add debug statements to the Bindings that can be activated with a `UTP_DEBUG_LOGGING` compile time option.
- Add [`debug`](https://github.com/visionmedia/debug) statements to the Connection and UTP classes that also let know what happens when.
- (Optional) Add debug builds of the binaries and load the debug builds if the `debug.enabled` is true instead of the production binaries.

Start with the failure path around utp_internal.cpp line 2181 and inspect the bindings plus the UTP and Connection classes. Trace how connection state changes reach the bindings, then define the debug logging and connection identifiers across those layers. Done means the requested compile-time bindings logging and class-level debug output are available, with optional debug binaries assessed separately.

Written by the indexing model from the issue text.