Cleanup of barcode.codex._convert

[maresb]

Most invocations of _convert are guaranteed to not handle digits in charset C. Handling extra digits in charset C introduces extra complexity due to grouping of pairs of digits, which is handled by _buffer. (I rename _buffer to _digit_buffer for clarity.)

In order to isolate the complexity and ease type hinting, I restrict _convert to return only int and raise an informative exception whenever the buffer is triggered. Correspondingly I introduce a function _convert_or_buffer returning int | None to handle the full case when _digit_buffer might be used.

There is just one single place (_build) where _convert_or_buffer is necessary rather than _convert.

Reasons why _convert is safe to use in all other invocations:

• In _new_charset it's used to switch charset, so it won't be used on a digit.
• In _maybe_switch_charset and _build it's used to flush the buffer immediately after switching to either charset A or B, so the charset cannot be C.
• In _convert_or_buffer I explicitly check before invoking that the charset isn't C.

This makes it clear why only the invocation in _build needs to handle the None return type. (IMO it's pretty tricky to deduce this if you're not already familiar with the implementation.)