guard shrink-factor division against floating-point near-zero

[harsha-cpp]

fixes #1665.

problem: when all flex children are frozen to their min-width in the first
pass, totalFlexShrinkScaledFactors is reduced to near-zero by floating-point
cancellation rather than exactly 0. the guard in distributeFreeSpaceSecondPass
uses exact equality (== 0), so it never fires, and the second pass divides
remainingFreeSpace by a value on the order of 1e-7, producing a childSize on
the order of 1e11 that overwhelms the min/max clamp.

fix: replace the exact-zero check with a relative epsilon guard
(shrinkFactorMagnitude < 1e-6f). when the magnitude is that small, all items
were already frozen in the first pass; the safe fallback (childFlexBasis + flexShrinkScaledFactor) applies and the subsequent boundAxisWithAutoMin clamps
correctly to minWidth.

regression: YGFlexShrinkBorderBug.flex_basis_0_border_minwidth_row reproduces
the original crash — 4 children, borderWidth difference of 1e-6 across them,
all now compute to their correct minWidth.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
yoga-website	Ready	Preview, Comment	Jun 9, 2026 4:37pm

[cipolleschi]

Nicely spotted! Thanks for the fix

[meta-codesync]

@cipolleschi has imported this pull request. If you are a Meta employee, you can view this in D108030908.

[harsha-cpp]

thanks @cipolleschi