Avoid false signed-out detection on MiniMax coding-plan shell pages

[qipihen]

Summary

• fix the existing MiniMax web fetch path so shell pages are not misclassified as signed out just because login strings appear inside script/style payloads
• strip non-visible HTML content before checking for sign in / log in / 登录 / 登入
• add a MiniMax web fetch regression test that exercises the fallback from a __NEXT_DATA__ shell page to the remains API

Scope

This PR does not add MiniMax support as a new provider.
It fixes the MiniMax web-fetch path that already exists in the current upstream codebase.

Why

MiniMax's current Coding Plan page can return an SSR shell where login-related translation strings live inside __NEXT_DATA__, even though the page is not actually the sign-in screen. The previous raw-string check saw 登录 in the script payload and returned invalidCredentials before the fetcher could fall back to coding_plan/remains.

Validation

• reproduced locally with a live MiniMax web session and confirmed the old logic returned MiniMax credentials are invalid or expired
• verified the patched local build successfully reports MiniMax usage again from the web source
• attempted swift test --disable-experimental-prebuilts --filter MiniMax on a clean local checkout; that run is currently blocked here by an unrelated KeyboardShortcuts #Preview macro build issue, not by this diff

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4448d5e4ed

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve script-only login detection for signed-out shells

Switching looksSignedOut to only scan visibleText drops all <script> content, so a JS-rendered sign-in shell with no visible login copy now returns false here even when auth has expired. In that case fetchCodingPlanHTML/fetchCodingPlanRemains continue into parse(html:), which can surface parseFailed (or partial script-derived data) instead of invalidCredentials, changing user-facing behavior for expired cookies. Consider keeping a fallback raw-HTML login signal when extracted visible text is empty.

Useful? React with 👍 / 👎.

[ratulsarna]

Thanks for PR @qipihen !

Small concern: since looksSignedOut now scans only visibleText, a real JS login shell with no visible copy might slip through and become parseFailed instead of invalidCredentials.

Maybe keep a raw-HTML fallback when visible text is empty?