The overhead of str::push_char is high enough to cripple the performance
of these two functions. I've switched them to build the output in a
~[u8] and then convert to a string later. Since we know exactly the
bytes going into the vector, we can use the unsafe version to avoid the
is_utf8 check.
I could have riced it further with vec::raw::get, but it only added
~10MB/s so I didn't think it was worth it. ToHex is still ~30% slower
than FromHex, which is puzzling.
Before:
```
test base64::test::from_base64 ... bench: 1000 ns/iter (+/- 349) = 204 MB/s
test base64::test::to_base64 ... bench: 2390 ns/iter (+/- 1130) = 63 MB/s
...
test hex::tests::bench_from_hex ... bench: 884 ns/iter (+/- 220) = 341 MB/s
test hex::tests::bench_to_hex ... bench: 2453 ns/iter (+/- 919) = 61 MB/s
```
After:
```
test base64::test::from_base64 ... bench: 1271 ns/iter (+/- 600) = 160 MB/s
test base64::test::to_base64 ... bench: 759 ns/iter (+/- 286) = 198 MB/s
...
test hex::tests::bench_from_hex ... bench: 875 ns/iter (+/- 377) = 345 MB/s
test hex::tests::bench_to_hex ... bench: 593 ns/iter (+/- 240) = 254 MB/s
```
FromHex ignores whitespace and parses either upper or lower case hex
digits. ToHex outputs lower case hex digits with no whitespace. Unlike
ToBase64, ToHex doesn't allow you to configure the output format. I
don't feel that it's super useful in this case.
