Handle chunked multibyte characters

[dmarkow]

In some large messages (in my case, roughly 8MB strings with very large JSON objects) I was running into JSON.parse failures on the server side.

Before this fix, if the split between different chunks in a message occurred in the middle of a two-byte string, the first byte would still be parsed by data.toString() and converted into a � character. Then the second byte in the next chunk would also be converted into a � character. This would then throw the length of the string off increasing it by 1, and since this library uses the content length prefix at the beginning of the string, it would effectively ignore the last character, which is often a } or ]. So JSON.parse would fail because of the missing closing bracket.

Node's built-in StringDecoder module is made specifically to ensure decoded strings don't contain incomplete multibyte characters. When it tries to decode a buffer that has an incomplete character, it'll set that last byte or two aside and hold on to them until the next time it's called.

I believe this will address the concerns in #11.

[SamuelBolduc]

Thank you for your PR. I personally encountered problems in the past that were most likely related to this issue.

For some reason Travis has a hard time running the tests on the pull-request, but I validated the fix and all the tests, including the new one submitted with the commit, pass without any issue. I also tested the code in real-world situation and it works flawlessly.