According to the discussion on golang.org/cl/55210,
adding benchmarks for reading from and writing to tar archives.

Splitting the benchmarks into 3 sections of USTAR, GNU, PAX each.

Results ran with -cpu=1 -count=10 on an amd64 machine (i5-5200U CPU @ 2.20GHz)
name           time/op
/Writer/USTAR  5.31µs ± 0%
/Writer/GNU    5.01µs ± 1%
/Writer/PAX    11.0µs ± 2%
/Reader/USTAR  3.22µs ± 1%
/Reader/GNU    3.04µs ± 1%
/Reader/PAX    7.48µs ± 1%

name           alloc/op
/Writer/USTAR  1.20kB ± 0%
/Writer/GNU    1.15kB ± 0%
/Writer/PAX    2.61kB ± 0%
/Reader/USTAR  1.38kB ± 0%
/Reader/GNU    1.35kB ± 0%
/Reader/PAX    4.91kB ± 0%

name           allocs/op
/Writer/USTAR    53.0 ± 0%
/Writer/GNU      47.0 ± 0%
/Writer/PAX       107 ± 0%
/Reader/USTAR    32.0 ± 0%
/Reader/GNU      30.0 ± 0%
/Reader/PAX      67.0 ± 0%

Change-Id: I58b1b85b52e58cbd566736aae4d722a3ddf2395b
Reviewed-on: https://go-review.googlesource.com/55254Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Run-TryBot: Joe Tsai <thebrokentoaster@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>