Text to integer fast path for small integers?

[oscbyspro]

I wrote some code for a comment on the forums. In doing so, I noticed two things:

1. Parsing up to the first invalid character without failing entirely by returning nil might be useful.
2. I may speed up small integer decoding by removing the arbitrary precision overhead for types that fit in one word.

This issue tracks the 2nd point. I'll want to measure it, but it should be a simple improvement.

[oscbyspro]

So this is a bit silly. Decoding an unsigned machine word with wrapping behavior is about 50% faster than decoding with error handling. The current arbitrary precision algorithm uses the wrapping version internally, so it ends up being about the same speed as a naive small integer algorithm with error handling. The former gets away with wrapping behavior because of its chunking strategy. Perhaps I should do something similar in the small integer algorithm? Hm.