performance: reverse example artifact runs much slower compared to plain

[gaukas]

goos: linux
goarch: amd64
pkg: github.com/refraction-networking/water/transport/v0
cpu: 12th Gen Intel(R) Core(TM) i5-1240P
BenchmarkDialerOutbound-16                176220              5844 ns/op         175.23 MB/s
BenchmarkDialerOutboundReverse-16          12115             89924 ns/op          11.39 MB/s
BenchmarkInboundListener-16               209980              5848 ns/op         175.11 MB/s
BenchmarkInboundListenerReverse-16         10000            135864 ns/op           7.54 MB/s
PASS
ok      github.com/refraction-networking/water/transport/v0     7.072s

More investigation required, preliminary guesses are pointing towards dynamic allocation/de-allocation.