…22402)

Stacked on #22400. Closes the second `cjs-module-lexer` gap.

After #22400, `Object.defineProperty(exports, "name", ...)` is fine, but `exports["name"] = …` / `module.exports["name"] = …` were still being emitted with backtick keys in minify. The lexer recognises this form too — it just couldn't see them past the template literal.

Valid-identifier keys (`exports["foo"]`) are already collapsed to dot access by the minifier earlier in the pipeline, so only non-identifier keys (`exports["has-dash"]`, `module.exports["__esModule"]`, …) survive to the codegen stage and were affected.

In `AssignmentExpression::gen_expr`, detect when the LHS is a computed member expression on `exports` / `module.exports` whose key is a `StringLiteral`, and print the key via `print_string_literal(key, false)`. Other computed access (`otherObj["x"]`) is unaffected.

### Where each cjs-module-lexer named-export form lands

| Pattern | Status |
|---|---|
| `exports.name = …` | identifier — never affected |
| `module.exports.name = …` | identifier — never affected |
| `module.exports = { "name": … }` | object property keys already use `allow_backtick: false` |
| `Object.defineProperty(exports, "name", …)` | fixed in #22400 |
| `exports["name"] = …` / `module.exports["name"] = …` | **fixed here** |

### Before / after

```js
// input
exports["has-dash"] = a;
module.exports["__esModule"] = true;
otherObj["not-exports"] = d;
```

```js
// before
exports[`has-dash`]=a,module.exports[`__esModule`]=!0,otherObj[`not-exports`]=d;

// after
exports["has-dash"]=a,module.exports.__esModule=!0,otherObj[`not-exports`]=d;
```

(`module.exports.__esModule` collapses to dot via a separate minifier pass; `otherObj["x"]` correctly stays backtick.)

Refs #22342

AI assisted.