Fine-grained child insert/remove invalidation based on ElementSelectorFlags

[nicoburns]

Summary

Replaces the blanket parent RestyleHint::restyle_subtree() on every child insertion/removal (and the TODO: make this fine grained / conditional based on ElementSelectorFlags comments) with flag-driven invalidation mirroring Gecko's RestyleManager::RestyleForInsertOrChange / ContentAppended / ContentWillBeRemoved, and fixes two invalidation gaps the new tests exposed.

New BaseDocument::restyle_for_child_insert_or_remove(parent_id, change_idx) (called after the child list is updated, with the insertion index or the removed child's old index):

• no relevant flags on the container → no restyles at all
• HAS_EMPTY_SELECTOR → restyle_for_empty_change: restyle the container's subtree, plus the container's later siblings when the grandparent has HAS_SLOW_SELECTOR_LATER_SIBLINGS (:empty + E / :empty ~ E), mirroring Gecko's RestyleForEmptyChange
• HAS_SLOW_SELECTOR (:nth-last-child family) → single restyle_subtree() on the container, like Gecko's RestyleWholeContainer
• HAS_SLOW_SELECTOR_LATER_SIBLINGS (:nth-child family) → restyle only element children at/after change_idx; a change at index 0 escalates to a whole-container restyle to avoid quadratic behavior when clearing a container from the start (Gecko's ContentWillBeRemoved optimization)
• HAS_EDGE_CHILD_SELECTOR (:first/last/only-child) → restyle the first/last element children and the nearest element children on either side of the change (covers the previously-last child on append / previously-first on prepend, like Gecko's MaybeRestyleForEdgeChildChange)

Repeated mutations within one style flush are deduped: if the container already carries a pending restyle_subtree() hint, the helper early-outs (the equivalent of Gecko's mRestyledAsWholeContainer set — Blitz can use the hint itself since it lives on the node and is cleared by the traversal).

All DocumentMutator child-list mutation paths route through it (add_children_to_parent, remove_node, remove_and_drop_node_with, remove_and_drop_all_children), which also fixes a pre-existing gap where remove_node inserted damage but never posted a restyle hint, so removals could leave stale sibling styles. The blanket parent restyle in set_attribute is removed entirely: attribute changes can't shift sibling indices, and stylo's snapshot invalidation covers attribute-dependent selectors (matching Gecko, which does no container-level work for attribute changes).

Two further fixes found while verifying against Gecko:

• Nodes moved within the document (old and new parent both in-document) now get restyle_subtree() themselves — previously masked by the blanket parent restyle, a moved node kept styles matched against its old ancestors (e.g. .red span).
• :empty text handling: set_node_text / append_text_to_node now trigger the empty-change restyle when a child text node transitions empty↔non-empty (Gecko's CharacterDataChanged path), and is_empty now ignores zero-length text nodes and comments per spec instead of treating any child as significant.

Known conservative divergence from Gecko (correctness-preserving): the :empty path restyles unconditionally when the flag is set rather than checking was/is-empty first.

Includes a new child_invalidation.rs test suite (7 tests) covering :nth-child, :nth-last-child, :last-child, :empty (structural, sibling-combinator, and text-content cases) and moved-node restyling.

Extracted from #800 so it can land first; the :nth-child(An+B of S) work will be rebased on top.

WPT results

No changes in test results compared to main.

_{Generated by the WPT workflow.}

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/65287500bfb04b2589b9fc6a08bc2d14
Open in Devin Desktop: https://dioxus.staging.devinenterprise.com/desktop/session/65287500bfb04b2589b9fc6a08bc2d14?variant=devin-insiders
Requested by: @nicoburns

[staging-devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring