Preserve complete UTF-8 characters in parse error snippets

[ydah]

Error snippets always drop their final multibyte character, even when it is complete.

require "json"

JSON.parse('[1, é')
# Before: unexpected character: '' at line 1 column 5
# After: unexpected character: 'é' at line 1 column 5

Only trim an incomplete character when the 32-byte limit splits it. Add coverage for complete characters, whitespace boundaries, and truncated 2-, 3-, and 4-byte characters.

[byroot]

Thank you!