Implement better buffer boundary matching algorithm

[soareschen]

Currently the multipart stream handler uses a very naive algorithm for matching boundary from multiple slices of buffers.

We want to avoid concatenating the buffers into single buffer because then it would be too large for finding the end boundary. This is unless io.js concatenates buffers without allocating and copying content into a contiguous concatenated buffer, which I haven't verify.

Until then, we need to consider using more efficient boundary matching algorithms such as the Boyer–Moore string search algorithm.