from ZXing (props Alex Geller) - about 25-60% slower depending on
data & stack heavy but does improve some outcomes when FNC1s
present (GS1_MODE or manual) although not much else it appears
(the previous algorithm was very good), but has a logical clarity
the other lacked - funcs `c128_dxsmooth()` etc shared with
CODE16K now moved there and renamed as `c16k_`;
also fix extended char latching when exactly 4 extended chars
at end
also manual code set switching now honoured exactly even if
immediate shift required;
manual: make explicit that AI "(00)" and "(01)" prefixes added by
Zint are HRT-only
general: add `extern "C"` wrappers to a few header files
93c3e27fba