Do not re-set container children that resolve to the same widgets

[maartenbreddels]

Why

The fast renderer skips the update of a widget when its element is the same object as in the last render and its kwargs resolve to the same values. But it compared the resolved kwargs (elements replaced by widgets) with the element kwargs (still elements). For a container this compare is always false, so every walked container got its children set again on every update.

For example, one leaf update next to 300 rows re-set the children of the VBox. One root update re-set the children of all 300 HBoxes.

What

For an unchanged element whose kwargs hold elements (a container, or layout=, v_slots=), we compare the resolved kwargs with the values the widget holds now. If they are the same objects, setting them would be a no-op, so we skip it. This change only affects the fast renderer.

A first version compared with the kwargs we set last time, kept in a dict per container. Review found that it changed behavior: a value changed from the frontend was no longer set back. Comparing with the live values keeps the old behavior, and needs no cache that could go stale.

Widget updates with real widgets, 300 rows (fast renderer):

	master	this PR
leaf update	2 (Button, VBox)	1 (Button)
root update	302 (300 HBox, VBox, Label)	2 (VBox, Label)

Comparing the 302 children with the live value costs ~33 µs. Setting them costs ~670–1900 µs (the timings were measured on a busy machine).

Tests

• test_leaf_update_does_not_update_sibling_containers: fails on master (fast renderer).
• test_same_element_sets_back_changes_from_outside: a value or children changed from outside is still set back. It passes on master and failed on the first version of this PR.
• Guard tests, both renderers: a child widget that changes type, a fragment that grows or shrinks, a fragment whose items are replaced at the same length, and a v_slots child that changes. I broke the compare on purpose (dicts always "same", lists compared by length only), and these tests catch it.
• Solara's unit suite gives the same results as with 1.10.3, with REACTON_FAST=0 and =1.

From #58 (its second commit), split off so it can ship on its own.

🤖 Generated with Claude Code