if rxq is buffered it adds only ~50ns to SyncChanRTT (was ~ 350ns ->
400ns) and important: no extra goroutines switches are introduced
(verified via analyzing trace).

50aa58a7d47d168ec550762c25d7aaaed3f4177b   ; oid=0..16995  nread=68269354  t=638.903389ms (37.591µs / object)  x=zsha1.go

No big speedup measurable but very visible in tracing.

before:

BenchmarkLinkNetPipeRTT-4         200000              9763 ns/op            9664 B/op         25 allocs/op
BenchmarkLinkTCPRTT-4             100000             19069 ns/op            9664 B/op         25 allocs/op

after:

BenchmarkLinkNetPipeRTT-4         200000              7872 ns/op            1224 B/op         17 allocs/op
BenchmarkLinkTCPRTT-4             100000             18465 ns/op            1221 B/op         17 allocs/op