Phase 8: StyleCache memoization

Add StyleCache struct with hash-keyed LRU-eviction strategy and
integrate into StyleSheet via Mutex<StyleCache> so no signature
changes are needed on evaluate_vdom_incr.

- StyleCache::get_or_compute hashes (type_name, props, epoch) -> ComputedStyle
- LRU: full clear on overflow (max 1024)
- RefCell avoided in favor of Mutex so StyleSheet remains Sync
  for parallel matching_rules_batch
- Manual Clone impl skips cache (fresh cache on clone)
- Call sites in evaluate_vdom_incr switched from ComputedStyle::compute
  to stylesheet.compute_cached
- Cache cleared in build_index after epoch increment (stale entries)

src/interpreter/mod.rs

@@ -469,7 +469,7 @@ impl Interpreter {
                         }
 
                         let matched_sheets = Self::collect_matching_styles(&vcomp, stylesheet, &[], &structural, ancestors, &sibling_infos[0..i]);
-                        vcomp.computed_style = ComputedStyle::compute(&vcomp.properties, &matched_sheets);
+                        vcomp.computed_style = stylesheet.compute_cached(vcomp.type_name, &vcomp.properties, &matched_sheets);
 
                         let child_ancestors = Self::build_ancestor_chain(ancestors, &vcomp);
                         let comp_child_refs: Vec<&Element<'a>> = comp.children.iter().collect();
@@ -502,7 +502,7 @@ impl Interpreter {
                         }
 
                         let matched_sheets = Self::collect_matching_styles(&vnode, stylesheet, &[], &structural, ancestors, &sibling_infos[0..i]);
-                        vnode.computed_style = ComputedStyle::compute(&vnode.properties, &matched_sheets);
+                        vnode.computed_style = stylesheet.compute_cached(vnode.type_name, &vnode.properties, &matched_sheets);
 
                         let child_ancestors = Self::build_ancestor_chain(ancestors, &vnode);
                         let child_refs: Vec<&Element<'a>> = el.children.iter().collect();