Convert escapeMetricName to use strings.Builder

[claytono]

This converts escapeMetricName to use strings.Builder instead of allocating a byte array, filling it and then converting it to a string. This also optimizes for the case where the metricName is already valid, and in that case, it just returns the original string.

This reduces memory allocations from 2 per call to 1 per call in the case when the string does need to be escaped, and reduces it to zero memory allocations when the string is already valid.

As discussed in #197, escapeMetricName is called a lot, and even after the improvements there, we still see this pretty far up the list in terms of cpu usage in profiling.

Benchmark before:

$ go test -benchmem -run=^$ github.com/prometheus/statsd_exporter -bench '^(BenchmarkEscapeMetricName)$' -benchtime 30s
goos: darwin
goarch: amd64
pkg: github.com/prometheus/statsd_exporter
BenchmarkEscapeMetricName/clean-8                       2000000000              33.6 ns/op            10 B/op          2 allocs/op
BenchmarkEscapeMetricName/0starts_with_digit-8          500000000               85.9 ns/op            64 B/op          2 allocs/op
BenchmarkEscapeMetricName/with_underscore-8             1000000000              55.7 ns/op            32 B/op          2 allocs/op
BenchmarkEscapeMetricName/with.dot-8                    1000000000              40.2 ns/op            16 B/op          2 allocs/op
BenchmarkEscapeMetricName/with😱emoji-8                 1000000000              57.4 ns/op            32 B/op          2 allocs/op
BenchmarkEscapeMetricName/with.*.multiple-8             1000000000              58.5 ns/op            32 B/op          2 allocs/op
BenchmarkEscapeMetricName/test.web-server.foo.bar-8     500000000               74.3 ns/op            64 B/op          2 allocs/op
BenchmarkEscapeMetricName/#00-8                         5000000000               9.59 ns/op            0 B/op          0 allocs/op
PASS
ok      github.com/prometheus/statsd_exporter   449.792s

Benchmark after:

go test -benchmem -run=^$ github.com/prometheus/statsd_exporter -bench '^(B
enchmarkEscapeMetricName)$' -benchtime 30s
goos: darwin
goarch: amd64
pkg: github.com/prometheus/statsd_exporter
BenchmarkEscapeMetricName/clean-8                       5000000000              8.42 ns/op            0 B/op          0 allocs/op
BenchmarkEscapeMetricName/0starts_with_digit-8          1000000000              45.0 ns/op            32 B/op          1 allocs/op
BenchmarkEscapeMetricName/with_underscore-8             3000000000              15.1 ns/op             0 B/op          0 allocs/op
BenchmarkEscapeMetricName/with.dot-8                    1000000000              37.3 ns/op             8 B/op          1 allocs/op
BenchmarkEscapeMetricName/with😱emoji-8                 1000000000              50.5 ns/op            16 B/op          1 allocs/op
BenchmarkEscapeMetricName/with.*.multiple-8             1000000000              58.9 ns/op            16 B/op          1 allocs/op
BenchmarkEscapeMetricName/test.web-server.foo.bar-8     500000000               75.3 ns/op            32 B/op          1 allocs/op
BenchmarkEscapeMetricName/#00-8                         10000000000             2.09 ns/op            0 B/op          0 allocs/op
PASS
ok      github.com/prometheus/statsd_exporter   367.326s

As you can see from the benchmarks, the common case where no escaping needs to be done is about 3-4x faster.

[matthiasr]

TIL about strings.Builder 😄 thank you!