Fix #1012: Prevent aggressive stripping of below/bellow header content wrappers

[hanumanraj07]

Fixes #1012

Description
Currently, Readability fails to parse articles on sites like knowablemagazine.org where the main content is wrapped inside a container with a class such as bellowheadercontainer.

The unlikelyCandidates regex aggressively strips out elements containing the word header without checking for word boundaries. As a result, structural containers using class names like below-header or bellowheader are prematurely removed from the DOM, causing Readability to completely miss the main article content and fallback to parsing unrelated footer elements.

This PR adds below and its common typo bellow to the okMaybeItsACandidate safeguard regex. This ensures that classes indicating placement beneath a header (which frequently wrap the main article content) bypass the unlikelyCandidates purge and receive a proper chance to be scored.

Testing

• All 1,984 automated tests pass successfully.
• Verified that okMaybeItsACandidate correctly safeguards the article container to resolve the parsing failure in knowablemagazine.org - Does not show the article, republish message displayed after enabling reader mode #1012.

[gijsk]

Thanks for the PR.

Unfortunately, it doesn't seem reasonable to treat "below" as semantically indicating "this is important article text" - even worse for "bellow".

Is there no other markup that can be used to make this determination? And/or, is there evidence that this class name structure is common for other sites than knowablemagazine.org, due to common CMSes or similar?

Also, please add a testcase to the testcases folder for this so we don't regress it in future - you can use the generate-testcase.js script to do this easily.

[gijsk]

Doesn't look like the PR changed since last time? I already left a comment before.

[hanumanraj07]

I'm closing this PR. After further investigation, adding below and bellow to okMaybeItsACandidate is too generic and doesn't semantically indicate the main article content.

We tried an alternative approach (adding word boundaries to header in unlikelyCandidates like \bheader\b) which correctly safeguards bellowheadercontainer, but this unfortunately broke several existing test cases (like the SRE book) where valid sub-headers were no longer stripped out.

Given that knowablemagazine.org hides its real content behind Cloudflare's CAPTCHA block and doesn't appear to have any other standard semantic markup on the article container, this seems to be a site-specific parsing failure that is not safely fixable on a global level. Thanks for your review!