fix(csv-stringify): quote fields fusing with a multi-character separator

[spokodev]

When a field is serialized before a multi-character delimiter (or record_delimiter), and the field’s tail is a non-empty prefix of that separator, the appended separator fuses with the field tail. On parse, the separator is then matched inside the original field, silently corrupting the data:

import { stringify } from "csv-stringify/sync";
import { parse } from "csv-parse/sync";

const s = stringify([["a:", "b"]], { delimiter: "::" }); // "a:::b\n"
parse(s, { delimiter: "::" });                           // [["a", ":b"]]  <- not [["a:", "b"]]

The same happens for a multi-character record delimiter with the last field of a record:

stringify([["x", "a#"]], { record_delimiter: "##" });    // x,a###...  -> parses as a record boundary

The quoting decision used value.indexOf(separator) >= 0, which quotes a field that contains the whole separator but misses this boundary fusion. Forcing quoted_string: true already produces correct, round-trippable output, confirming the field simply needs quoting.

The fix routes both delimiter and record_delimiter through a small emits_separator(value, separator) helper:

const emits_separator = function (value, separator) {
  return (
    separator.length !== 0 &&
    (value.indexOf(separator) !== -1 ||
      (separator.length > 1 &&
        (value + separator).indexOf(separator) < value.length))
  );
};

The boundary check runs only for multi-character separators, so single-character delimiters and non-fusing fields are unaffected (no extra quoting) — e.g. delimiter ":x" with field "a:" stays unquoted, since ":x" never starts inside "a:".

Tests added in option.delimiter.js and option.record_delimiter.js (fusing field is quoted; non-fusing field is not). Full suite passes (the one unrelated pre-existing failure in api.callback.js, "catch error in end handler #386", is present on master as well). dist/ is left for the release build, matching #484.