In-place and batch decode functions

[Krastanov]

In #212 we provide a number of decoders, however the current interface has performance detriments:

• we need to have an in-place decode! method that does not allocate a new array each time it outputs a guess
• we need a (in-place) bach decode_batch method that can take advantage of vectorization and SIMD (or at least it can take advantage of multi-threading)
• the current vcat used in the decoders is pretty bad for performance and a good first place to try a fix