Fix incorrect Mesh2d world normal and tangent computation

[kevthedawg]

Objective

• Fixes Mesh2d world_normal and world_tangent computation is incorrect #24614.
• mesh2d_normal_local_to_world and mesh2d_tangent_local_to_world don't normalize their result, so a scaled Transform scales world_normal by the inverse of the scale and world_tangent by the scale. examples/2d/mesh2d.rs scales by 128, giving a normal of length 1/128 and a tangent of 128.
• The tangent's w is passed through unchanged too, so mirrored transforms keep the wrong handedness. That is what makes this a breaking change.
• [Merged by Bors] - bevy_pbr: Fix tangent and normal normalization #5666 fixed both in bevy_pbr, but the mesh2d copies from [Merged by Bors] - Add reusable shader functions for transforming position/normal/tangent #4901 were left behind. SpriteMaterial calls the same functions and is affected too.

Solution

Bring both functions to parity with bevy_pbr: normalize, skipping zero vectors so they don't become NaN, and multiply the tangent's w by the sign of the model matrix determinant. That sign needs a SIGN_DETERMINANT_MODEL_3X3 flag on 2D's MeshFlags, so mesh2d_tangent_local_to_world now takes the instance index. The flag is derived in Mesh2dUniform::from_components, so custom extraction systems keep working unchanged.

Testing

New Tangents scene in examples/testbed/2d.rs draws the tangent space of a scaled quad and a mirrored one, turning red where a vector is not unit length. Removing either normalize(), or either half of the sign correction, turns the matching band red. No existing screenshot changes, since no built-in 2D fragment shader reads these, but the new scene needs a baseline.

[github-actions]

Welcome, new contributor!

Please make sure you've read our contributing guide, as well as our policy regarding AI usage, and we look forward to reviewing your pull request shortly ✨