Optimize some string handling

[moorereason]

I've added an additional commit to this PR that further optimizes this area of code:

benchmark                       old ns/op     new ns/op     delta
BenchmarkTreeToTomlString-2     49028         18817         -61.62%

benchmark                       old allocs     new allocs     delta
BenchmarkTreeToTomlString-2     485            78             -83.92%

benchmark                       old bytes     new bytes     delta
BenchmarkTreeToTomlString-2     9749          4961          -49.11%

[kevinburke]

You could probably speed this up even further by creating a bytes.Buffer outside of the loop range, writing to it inside the for loop, then calling w.Write(buf.Bytes()), which would also skip the string conversion below this line.

[moorereason]

Tried your suggestion, but it only added allocs. Perhaps I'm doing it wrong.

[kevinburke]

Here's what I had in mind. It makes the code a lot simpler, since we know that b.WriteString and company never return an error, we don't need to check them constantly.

b49b944

[pelletier]

@moorereason nice!
@kevinburke I like your proposition (makes the code much cleaner), but the +14% allocs increase isn't great. What happens with a larger preallocated buffer?

I think I'm going to merge this as a first step. Then a second step we should figure a way to make Kevin's solution use some less allocs, as it makes the code much more readable.

[pelletier]

Ref #167