Use pooled buffer in replaceShortcodes
Even as a copy at the end is needed, this consumes way less memory on Go 1.4.2:
```benchmark                           old ns/op     new ns/op     delta
BenchmarkParsePage                  145979        139964        -4.12%
BenchmarkReplaceShortcodeTokens     633574        631946        -0.26%
BenchmarkShortcodeLexer             195842        187938        -4.04%
benchmark                           old allocs     new allocs     delta
BenchmarkParsePage                  87             87             +0.00%
BenchmarkReplaceShortcodeTokens     9424           9415           -0.10%
BenchmarkShortcodeLexer             274            274            +0.00%
benchmark                           old bytes     new bytes     delta
BenchmarkParsePage                  141830        141830        +0.00%
BenchmarkReplaceShortcodeTokens     35219         25385         -27.92%
BenchmarkShortcodeLexer             30178         30177         -0.00%
```
See #1148