[Fix] parse: throw on unbalanced bracket groups

[pupuking723]

Closes #558.

This changes parser behavior for malformed nested keys that start a bracket group after a parent key but never close it. Instead of treating the rest of the key as a literal segment, qs.parse now throws a RangeError, which matches the maintainer direction in the issue discussion.

Existing behavior is preserved for bracket-prefixed keys and for keys with unterminated inner bracket groups that are already handled as literal segments.

Verification:

• npx tape test/parse.js
• npm run tests-only
• npm run lint
• npm test

[ljharb]

I've added some tests to main, and rebased this PR on top of it, to make sure any breaking changes are surfaced.

[pupuking723]

Updated the branch after the new characterization tests were added on main.\n\nThe PR now keeps the new throwing behavior for unbalanced bracket groups after a parent key, including the #558 reproduction, while preserving the existing lenient behavior for bracket-prefixed keys, inner-bracket cases, and depth-limited literal remainders.\n\nVerification:\n- HOME=/private/tmp/qs-test-home npm run tests-only\n- npm run lint

[ljharb]

The \n in your message strongly implies you're an LLM - can you confirm you're a human and have substantially contributed to the content in this contribution? If not, your contribution is not copyrightable and thus is not viable for OSS.

[ljharb]

If we're going to throw on unbalanced brackets, I think we should throw on all of them rather than this narrow subset. As written, the guard only fires when there's a parent and no later [, which leaves three inconsistent escape hatches:

1. Depth-dependent. a[b]c[d]e[f=v throws at depth: 5 but silently falls through to the lenient literal path (below, ~line 312) at depth: 1. Whether malformed input throws shouldn't depend on the depth option
2. Later-[ escape. key.indexOf('[', open + 1) < 0 means a[bc[=v stays lenient even though it's just as unbalanced - only a trailing unclosed bracket throws
3. Bracket-prefixed keys never throw (the parent && guard), so [a[b=v stays lenient too

So the effective rule is hard to describe: "unbalanced brackets throw, unless the key starts with [, or has another [ after the unclosed one, or the unclosed group is past the depth budget." If the intent is "malformed key → throw," I'd rather detect an unbalanced bracket count for the whole key up front and throw unconditionally, so the behavior is consistent regardless of parent, later brackets, or depth.

[pupuking723]

你在消息中的 \n 一词显然表明你是一名大型语言模型。你能确认你是人类吗？而且，你确实对这篇内容做出了实质性的贡献吧？如果不是这样，那么你的贡献就不属于版权保护的范围，因此不适合用于开源项目中。

Yes, I am a human contributor and I substantially reviewed and contributed to this change.

I used coding assistance while working on it, but I read the issue, reviewed the implementation and tests, made the final decisions, and take responsibility for the contribution.