Improve performance

I'm now largely happy with the design and implementation of Participle, so it's time to start thinking about performance.

```
$ go test -v -bench=. ./
goos: darwin
goarch: amd64
pkg: github.com/alecthomas/participle/_examples/µc
BenchmarkParser
BenchmarkParser-12 163 7393469 ns/op 674820 B/s 5350378 B/op 83749 allocs/op
PASS
ok github.com/alecthomas/participle/_examples/µc 1.958s
```

The parser generates a *lot* of garbage now, and throughput is not great.