Compatibility with json5 2.2.3

I tested tinyjson5 against json5 2.2.3 five ways:

json5's own parse, stringify, and errors tests, copied unchanged into test/json5 and run with tinyjson5 in place of json5. All 160 assertions pass: 49 parse, 67 stringify, 44 errors.
The json5-tests corpus. All 114 files are right: every .json and .json5 file parses, every .js and .txt file throws, and each gives json5's value or json5's error.
1,000,000 inputs made by editing corpus files, json5's test inputs, and generated documents at random. Every one gives the same value, or the same error message, lineNumber, and columnNumber (791,888 of 791,888 errors identical). The console.warn calls match too.
10,000 random values, with nested objects and arrays, awkward keys, both quote kinds, control characters, U+2028, NaN, -0, dates, boxed primitives, toJSON and toJSON5, functions, symbols, and cycles. Each is stringified with 6 space settings, 3 quote settings, and 3 replacers, positionally and as an options object, plus odd options on every tenth value: 741,000 calls with byte-identical output. Where json5 throws, in 9,993 of them, tinyjson5 throws the same TypeError: for cycles, and when a "__proto__" entry in the key list reaches Date.prototype.toJSON.
The same values round-trip: parse(stringify(x)) equals what JSON5 can hold of x in all 147,015 checks.

Where tinyjson5 differs:

Identifier characters. json5 2.2.3 ships Unicode 10 tables; tinyjson5 asks the JavaScript engine with \p{ID_Start} and \p{ID_Continue}. On Node 26.13 (Unicode 17) that adds 20,588 code points as an identifier start and 21,146 as an identifier part, such as U+0560 (added in Unicode 11) and U+00B7 MIDDLE DOT (an identifier part that json5's table leaves out). parse accepts unquoted keys with them where json5 throws, and stringify writes such keys bare where json5 quotes them. Older engines know fewer characters. json5 accepts no identifier character that tinyjson5 rejects; U+2E2F VERTICAL TILDE, which json5's tables count as a letter, is special-cased to keep it that way. pnpm unicode prints the full list.
Stack depth. Both parse nesting of any depth without recursion. stringify and the reviver recurse in both, and run out of stack at slightly different depths: here stringify handles 3,906 nested arrays to json5's 3,672, and the reviver 5,675 nested objects to json5's 5,859.
Packaging. ESM only, with named exports next to the default. There is no CommonJS or browser build, no CLI, and no register or require hook. The types are the same signatures, written for ESM, plus exported Reviver, Replacer, and StringifyOptions.