Advance past the whole codepoint after a zero-width regex match - #3586

What / why

After a zero-width match in a global ("g") regex, f_match advanced the search start by one byte to avoid re-matching at the same position. Oniguruma runs in ONIG_ENCODING_UTF8, so on a multibyte character that +1 byte lands inside the next codepoint, and the next onig_search finds a spurious extra zero-width match at a duplicated offset — one bogus match per multibyte character. match, gsub, sub, splits, split/2 and scan were all affected on multibyte input:

$ echo '"aé"'     | jq -c '[match("";"g")|.offset]'   # before [0,1,2,2]   after [0,1,2]
$ echo '"héllo"'  | jq -c 'gsub("(?=.)";"-")'         # before "-h-é--l-l-o"   after "-h-é-l-l-o"
$ echo '"aé"'     | jq -c '[splits("")]'              # before ["","a","é","",""]   after ["","a","é",""]

Fix

Advance past the whole codepoint using jvp_utf8_decode_length (already used elsewhere in f_match) rather than a single byte, so the search start can never land inside a multibyte character. This is the multibyte tail of #2565 — the ASCII case it fixed (and its test) still pass.

Tests

Added multibyte regression cases to tests/onig.test (offsets over 2/3/4-byte chars incl. €/🚀, gsub, splits). They fail on the previous code and pass now. ./jq --run-tests tests/onig.test → 50/50; full make check green.

Fixes #3585