Optimize UTF8 encoding

[davidik1]

We have observed several workloads where UTF_8$Encoder.encodeLoop called from StatsDProcessor$ProcessingTask.writeBuilderToSendBuffer is very prominent in the profile. We also noticed that a lot of time was spent getting and putting the characters/bytes, i.e., StringCharBuffer.get and DirectByteBuffer.put (see screenshot).
This is actually quite similar to the grey frames in the screenshot in the following comment (although it dealt with a completely separate issue): #203 (comment)
After applying the changes suggested in this PR (replacing the CharsetEncoder interface with plain String.getBytes) performance improved dramatically and writeBuilderToSendBuffer is no longer visible in the profile.
We also found a pretty old blogpost benchmarking the two: https://www.evanjones.ca/software/java-string-encoding.html.

[YoniKF]

Hi @vickenty and team,
Is this change something you are willing to accept into the project?
We (Intel Granulate) have observed this inefficiency on several services of our clients and think that it makes sense to fix it at the source.