fix: update Taffy to 0.14.0

[hyfdev]

Both TaffyJS runtimes use Taffy 0.14.0, which prevents deeply nested flex columns from repeating the same layout work at each level and preserves preferred widths for auto-sized descendants beneath a capped ancestor. The JavaScript API and binding surface are unchanged.

For a root capped at 8 px around an auto-width middle node and a leaf with width: 86:

[root, middle, leaf].map((node) => tree.getUnroundedLayout(node).size.width);
// [8, 86, 86]

Problem

At Taffy revision 8d13fdc8, the per-node measurement cache maps different parent-size requests to the same fixed entry, so nested column containers repeatedly evict work they need again. DioxusLabs/taffy#1148 tracks that performance defect.

The same revision also reuses one cached measurement for two different flex sizing questions. A preferred width can therefore disappear when the descendant should size to its content and overflow a capped ancestor. DioxusLabs/taffy#1149 tracks that correctness defect.

Both the native and Wasm packages compile the same Rust binding, so both exposed these behaviors.

Result

• Taffy is pinned to 77f38568, the Taffy 0.14.0 release commit. It includes the new multi-entry measurement cache, applies preferred flex sizes separately from content measurement, and also contains upstream intrinsic-sizing fixes for cleared floats and negative margins.
• The shared public API suite covers the three-node preferred-width case against both @taffyjs/node and @taffyjs/wasm.
• The lockfile and internal evidence links retain the exact source revision, while public package and website docs identify the release as Taffy 0.14.0.
• The published benchmark dataset contains a complete run from its fixed Apple M3 Pro host.

Benchmarks

The publication profile puts @taffyjs/node at parity with or ahead of yoga-layout in nine of twelve comparable scenarios under #53's 1.10× parity boundary. The cache-sensitive and remaining measured scenarios are:

Scenario	@taffyjs/node	yoga-layout	Relative time	Measure calls, node / Yoga
Incremental frame: one leaf	0.235 ms	0.436 ms	+1.86×	5 / 1
Incremental frame: one tenth	3.771 ms	0.591 ms	-6.39×	785 / 10
Incremental frame: every leaf	4.798 ms	2.571 ms	-1.87×	1,074 / 87
Measured render: medium	1.610 ms	1.425 ms	-1.13×	440 / 37
Nesting depth: 2	0.484 ms	0.818 ms	+1.69×	0 / 0
Nesting depth: 6	0.539 ms	0.742 ms	+1.37×	0 / 0
Nesting depth: 10	0.575 ms	0.752 ms	+1.31×	0 / 0

The complete publication profile passed the repository's 10% relative-margin-of-error and 25% round-median-spread gates. The remaining three slower scenarios all measure text through JavaScript callbacks and remain tracked in #53.

Validation

• vp run check
• vp run check:wasm
• vp run check:codegen
• vp run benchmark:update-website
• vp run build:website

The shared preferred-width regression is red at Taffy revision 8d13fdc8 with [8, 8, 8] and green at Taffy 0.14.0 with [8, 86, 86]; check:wasm runs the same suite against the Wasm package.

Part of #53.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.