Use llvm::DenseSet for seadsa_heap_abs_impl::NodeSet

[LinerSu]

Summary

• Replace seadsa_heap_abs_impl::NodeSet backing storage from std::set to llvm::DenseSet.
• Add orderedNodes to preserve deterministic node traversal where read/mod/new and equivalence-class results are built.
• Update NodeSet union and difference helpers to use DenseSet insert and erase operations.

Why

std::set is inefficient for this use case because the node collections are primarily membership sets, so the tree ordering overhead is unnecessary. llvm::DenseSet is a better fit for lookup, insertion, and deletion. Where iteration order affects observable output, nodes are sorted explicitly by SeaDsa node ID before traversal.

[caballa]

The original implementation of set_difference and set_union runs in O(|s1| + |s2|) because it assumes that s1 and s2 are sorted. Your implementation runs in O(|s2| x log(|s1|) which is slower for large sets. What is your motivation for changing them?

[caballa]

This code is equivalent but cleaner in my opinion

OrderedNodeVec nodes(set.begin(), set.end());
std::sort(nodes.begin(), nodes.end(), NodeOrdering{});
return nodes;