Improve additional field type discovery

[mp911de]

Right now, additional fields (string/double-typed) in discovery mode have a high cost of type discovery because states are communicated using exceptions.

A benchmark proves this cost:

Benchmark                                   Mode  Cnt     Score    Error  Units
GelfMessageBenchmark.configuredDoubleField  avgt    5    52,522 ±  1,848  ns/op
GelfMessageBenchmark.configuredLongField    avgt    5    36,982 ±  0,614  ns/op
GelfMessageBenchmark.configuredStringField  avgt    5    18,341 ±  0,324  ns/op
GelfMessageBenchmark.discoverDoubleField    avgt    5  1520,149 ± 88,000  ns/op
GelfMessageBenchmark.discoverLongField      avgt    5    25,537 ±  0,131  ns/op
GelfMessageBenchmark.discoverStringField    avgt    5  2714,702 ± 49,806  ns/op

[mp911de]

Improving discovery with own character scanning to discover the most appropriate type could help to improve parsing durations.

Benchmark                                   Mode  Cnt   Score   Error  Units
GelfMessageBenchmark.configuredDoubleField  avgt    5  53,310 ± 3,263  ns/op
GelfMessageBenchmark.configuredLongField    avgt    5  39,175 ± 1,899  ns/op
GelfMessageBenchmark.configuredStringField  avgt    5  19,136 ± 0,771  ns/op
GelfMessageBenchmark.discoverDoubleField    avgt    5  56,483 ± 4,365  ns/op
GelfMessageBenchmark.discoverLongField      avgt    5  29,863 ± 3,405  ns/op
GelfMessageBenchmark.discoverStringField    avgt    5   6,131 ± 0,481  ns/op