Current code uses trees of bytes.Buffer as data representation.
Each bytes.Buffer takes 4k bytes at least, so it's waste of memory.
The change introduces trees of lazy-encoder as
alternative one which reduce allocations.

name       old time/op    new time/op    delta
Marshal-4    64.7µs ± 2%    42.0µs ± 1%  -35.07%   (p=0.000 n=9+10)

name       old alloc/op   new alloc/op   delta
Marshal-4    35.1kB ± 0%     7.6kB ± 0%  -78.27%  (p=0.000 n=10+10)

name       old allocs/op  new allocs/op  delta
Marshal-4       503 ± 0%       293 ± 0%  -41.75%  (p=0.000 n=10+10)

Change-Id: I32b96c20b8df00414b282d69743d71a598a11336
Reviewed-on: https://go-review.googlesource.com/27030Reviewed-by: Adam Langley <agl@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Adam Langley <agl@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>