#2: Handle buffer size and other errors properly

[nwoods]

This is my first time working on an implementation of the codec API so please be patient if I did anything stupid.

Changes:

• If output buffer is too small, make it bigger when trying again
• Fix indexing confusion (msg is bytes, index returned by error can be used unmodified).
• Pass input to encode retries in correct format (and fix a few other typing issues).
• If decode fails due to an invalid sequence with errors='ignore' or errors='replace', remove the offending byte before retrying.

When the buffer is too small, I increase it by 10% (min 3 bytes). I chose those numbers based on nothing whatsoever, and I'm open to other suggestions.

The unit tests succeed, and this works for my use cases plus some others I fiddled around with, but I wouldn't say it has been tested extensively.

[nwoods]

Edit: Factored implementations into separate functions to avoid unnecessary decode-encode during encoding retry when buffer is too small, and to avoid changing the interface of functions that should respect the codec API.