Call copy with as large buffer as possible to reduce the
number of function calls.

benchmark                 old ns/op    new ns/op    delta
BenchmarkBytesRepeat            540          162  -70.00%
BenchmarkStringsRepeat          563          177  -68.56%

LGTM=josharian
R=golang-codereviews, josharian, dave, dvyukov
CC=golang-codereviews
https://golang.org/cl/90550043