Performance improvements on vector/map parsing

[leiffoged]

Hey there --

I'm using clojure-msgpack with parsing a lot of vectors in a performance critical path of my application. Profiling revealed that the seq overhead involved with both the (for ...) and the (range ...) sequences is significant.

This diff eliminates the unnecessary allocation by:
1/ unrolling them into explicit loops w/ transients
2/ using unchecked math to avoid boxing the counter

One change to call out: unpack-n now returns a vector instead of a sequence. Happy to wrap in a call to sequence if that's something you're interested in maintaining (e.g. a real sequence instead of something sequential)

Tested by invoking lein test

[edma2]

LGTM, good catch!