…ible by 11

The `_calculate_checksum` method in `InternationalStandardSerialNumber`
was missing a final `% 11`, causing it to return 11 (an invalid
two-digit value) instead of 0 when the weighted digit sum happened to
be exactly divisible by 11.

Fixes WhyNotHugo#238.

In charset C the digit pair "99" encodes to the code number 99, which
is also the TO_C charset-switch marker. Code128._try_to_optimize folded
away a leading switch code, so a barcode whose data began with "99" had
that pair silently dropped: Code128("9912345678").encoded produced
[105, 12, 34, 56, 78], decoding to "12345678".

Restrict the START-code folding to a genuine START_C followed by a
switch to charset A or B, so the data value 99 is preserved while the
legitimate optimization (e.g. "Wikipedia" starting in charset B) still
applies. Fixes WhyNotHugo#251.

Subclass digits overrode EuropeanArticleNumber13.digits, so get_fullcode()
returned a truncated barcode while str() still showed the correct ISBN/ISSN.

Code128._build mutated self._charset and self._digit_buffer (both set
only in __init__) but never reset them. After a build that switched to
charset A/B or left a digit buffered, a subsequent build()/encoded call
started from stale state and emitted a different, incorrect bit string.

Reset both to their initial values at the start of _build() so repeated
calls on the same instance are deterministic.

Fixes WhyNotHugo#143

Co-Authored-By: Hugo Osvaldo Barrera <[email protected]>