If the json.encoder code does get updated, it doesn't need two levels of matching.  It can be flattened by eliminating the *chunks* variable.

    match value:
        case str():
            yield _encoder(value)
        case None:
            yield 'null'
        case True:
            yield 'true'
        case False:
            yield 'false'
        case int():
            yield _intstr(value)
        case float():
            yield _floatstr(value)
        case list() | tuple():
            yield from _iterencode_list(value, _current_indent_level)
        case dict():
            yield from _iterencode_dict(value, _current_indent_level)
        case _:
            yield from _iterencode(value, _current_indent_level)