Slightly speed up scripts

[nik9000]

This saves a couple of nanoseconds on ever invocation of a painless
script that uses SearchLookup. Which is most of the scripts in the hot
path. This amounts to a 3.5% performance savings for things like script
scoring and runtime fields. We got this by speeding up a method that the
async profiler reported as taking 20% of the time running a script score
with painless. This drops it to 1%. You'd think that'd result in a 19%
speed up. But no such luck.

[elasticmachine]

Pinging @elastic/es-search (Team:Search)

[nik9000]

The actual speedup from this is pretty disappointing. The async profiler has betrayed me! But its a very very easy change so I figure its worth it.

[nik9000]

Benchmarks just finished:

Before:

Benchmark                       (indexingBufferMb)       (script)  Mode  Cnt   Score   Error  Units
ScriptScoreBenchmark.benchmark                  16     expression  avgt   10  63.950 ± 1.444  ns/op
ScriptScoreBenchmark.benchmark                  16          metal  avgt   10  52.474 ± 0.572  ns/op
ScriptScoreBenchmark.benchmark                  16  painless_cast  avgt   10  71.872 ± 0.472  ns/op
ScriptScoreBenchmark.benchmark                  16   painless_def  avgt   10  72.232 ± 1.007  ns/op

After:

Benchmark                       (indexingBufferMb)       (script)  Mode  Cnt   Score   Error  Units
ScriptScoreBenchmark.benchmark                  16     expression  avgt   10  63.203 ± 0.063  ns/op
ScriptScoreBenchmark.benchmark                  16          metal  avgt   10  52.388 ± 0.312  ns/op
ScriptScoreBenchmark.benchmark                  16  painless_cast  avgt   10  70.971 ± 0.235  ns/op
ScriptScoreBenchmark.benchmark                  16   painless_def  avgt   10  70.097 ± 0.469  ns/op

Like I said - its a small improvement for painless scripts. But its a small change.

[nik9000]

The async profiler led me to believe it'd be a bigger improvement or else I wouldn't even have looked. But still.

[romseygeek]

LGTM