Fix AIndex.entrySet() losing sorted key order via HashSet

[anutechrepo77]

Fixes #651.

Problem

AIndex documents itself as "a sorted radix-tree map of Blobs to Values" whose primary benefit is to "provide sorted orderings for indexes." That guarantee holds for entryAt(i), but not for entrySet():

@Override
public Set<Entry<K, V>> entrySet() {
    HashSet<Entry<K,V>> hs=new HashSet<>(size());
    long n=count();
    for (long i=0; i<n; i++) {
        MapEntry<K,V> me=entryAt(i);
        hs.add(me);
    }
    return Collections.unmodifiableSet(hs);
}

entryAt(i) walks the radix tree correctly in ascending key order, but the entries are collected into a java.util.HashSet, whose iteration order is hash-bucket-based, not insertion order. Any caller iterating entrySet() directly (or keySet(), forEach, an enhanced-for loop) — a reasonable thing to do given the class's documented sort guarantee — gets entries back in effectively arbitrary order instead.

Fix

Swap HashSet for LinkedHashSet. Since entryAt(i) already inserts in correct ascending order for i = 0..n-1, a LinkedHashSet (which preserves insertion order) is sufficient to restore the documented guarantee — no other logic changes.

Testing

Added IndexTest.testEntrySetOrder(), which builds a 50-entry Index and asserts entrySet()'s iteration order matches entryAt(i) for every i.

• Fails on the old HashSet implementation (confirmed locally: entry 0 expected, entry 49 observed at the same position)
• Passes with the LinkedHashSet fix
• Full convex-core suite: 2812 tests, 0 failures, 0 errors

---

🤖 Generated with Claude Code