fix: prevent browser freeze when tessellating >50k vertices (dev-2.0)

[Nixxx19]

Resolves #8219 for the dev-2.0 branch

This is the dev-2.0 port of #8555.

Summary

• Adds a one-time confirmation prompt before tessellating very large PATH-mode shapes (>50,000 vertices) to avoid browser freezes in libtess.
• Cancel: skips tessellation and draws nothing for that shape.
• OK: remembers the choice for the lifetime of the renderer via _largeTessellationAcknowledged and tessellates normally without prompting again.
• Respects p5.disableFriendlyErrors: when disabled, tessellation always runs with no prompt.

Tests

Added unit tests in test/unit/webgl/p5.RendererGL.js under the beginShape() in WEBGL mode suite, written in vitest syntax (vi.spyOn / mockImplementation). All 4 tests pass locally.

[Nixxx19]

[p5-bot]

Continuous Release

CDN link

• https://raw.esm.sh/pr/p5@5b1051e/lib/p5.min.js

Published Packages

• https://pkg.pr.new/p5@5b1051e

Commit hash: 746e46c

Previous deployments

---

This is an automated message.

[davepagurek]

Ah yes this is one of the currently kinda awkward bits of 2.x now that we've split core functionality into addons: there is not always a global p5. The way we're currently dealing with this is with a sort of "dependency injection" pattern of sorts, where we make a function on the current class, like in this case maybe friendlyErrorsDisabled() { return false; }, and then in an addon file inside of function myAddon(p5, fn, lifecycles) { ... }, we override that function with a real implementation. Take a look in this PR at how we've given p5.Image access to p5.friendlyError: https://github.com/processing/p5.js/pull/8676/changes#diff-fdf1c89a0511ed981f59e9539de1ff707da1e75c7b21f35184406cacb5781672

Let's maybe try to follow a similar pattern here where we override ShapeBuilder.friendlyErrorsDisabled (or whatever we choose to call it) from within the addon function that adds the WebGL renderer to p5?

[Nixxx19]

thanks for the pointer! yeah that pattern makes way more sense, the _pInst.constructor thing was definitely awkward. added a friendlyErrorsDisabled() on ShapeBuilder that returns false by default, and then overrode it inside the renderer3D addon where p5 is in scope to actually check p5.disableFriendlyErrors. tests still pass. let me know if you had something slightly different in mind!