fix: improve performance of encode and decode

[dwalluck]

Don't bother partitioning array
Skip bytes already in the decode buffer

[dwalluck]

Sorry for the over-engineering in the first place, but when I got around to running some actual JMH benchmarks, it was clear that we were losing performance likely due to having to read the array twice to try to write it back in chunks. So, just go back to writing one at a time.

There was a bug where decoding wrote the same bytes that were already there instead of just setting the position. For decode, we can continue to look for the first '%' since this doesn't require reading again, just starting at that position.