SOLR-18315: Optimize UTF conversions by using standard Java APIs

[psalagnac]

https://issues.apache.org/jira/browse/SOLR-18315

Description

This replaces class ByteUtils by Java API calls, mostly String constructor and getBytes() method when doing short-length byte[] <-> String conversions, or using CharsetEncoder for longer data.

Tests

I validated the perf improvement with RequestWriters benchmark (note it just measures serialization of update requests, whithout processing the requests).

I get a 8.5x improvement in throughput.

### BASELINE
Benchmark                   (batchSize)   (type)   Mode  Cnt     Score     Error  Units
RequestWriters.writeUpdate           10  javabin  thrpt    5  5660,213 ± 277,626  ops/s
RequestWriters.writeUpdate          100  javabin  thrpt    5   572,390 ±   3,854  ops/s
RequestWriters.writeUpdate         1000  javabin  thrpt    5    57,397 ±   0,270  ops/s
RequestWriters.writeUpdate        10000  javabin  thrpt    5     5,704 ±   0,039  ops/s

### AFTER
Benchmark                   (batchSize)   (type)   Mode  Cnt      Score      Error  Units
RequestWriters.writeUpdate           10  javabin  thrpt    5  49797,001 ± 2110,121  ops/s
RequestWriters.writeUpdate          100  javabin  thrpt    5   5245,920 ±  143,421  ops/s
RequestWriters.writeUpdate         1000  javabin  thrpt    5    514,228 ±   15,048  ops/s
RequestWriters.writeUpdate        10000  javabin  thrpt    5     48,803 ±    0,875  ops/s

[epugh]

Bug fix here? Nice to see the benchmarks get used and therefore finding oversights like this.