minimal protection against invalid UTF8 encoding.

[rdentato]

Accepts invalid encodings

• multiple representation for the same character accepted
• stray non-UTF8 bytes accepted

The minimal protection I've added is to ensure that it does not override the str_buffer memory if the number of stray continuations is greater than max_token_length*2+1

[rdentato]

I've also added some space to str_buffer in case max_token_length is one. Actually that one was needed anyway to allow str_buffer to be used for collecting UTF8 Enconding and not only for handling pairs of tokens.

[karpathy]

I like it ty

[rdentato]

@karpathy. I'll keep the validation branch in my repo and close the PR #300.
Should you feel the need to go through validation, I'll resume it.