Build Merkle trees directly from sorted leaves

[robknight]

This PR changes how Merkle trees are built from HashMaps. Previously, we would iterate over the HashMap, inserting each node into a full Merkle tree, repeatedly traversing existing paths and writing intermediate nodes that were superseded by later insertions. This is expensive when we have to build a Merkle tree containing thousands of nodes, e.g. when building a large custom predicate module.

Building from sorted leaves performs one pass over the final tree structure and stores each leaf and intermediate node exactly once, substantially reducing constructor work and database traffic.

• Replace repeated insertion in MerkleTree::new_with_db with direct bulk construction.
• Sort leaves by path and recursively construct the canonical tree.
• Store only nodes reachable from the final root, avoiding transient nodes created by incremental insertion.
• Add tests comparing roots, iteration results, and proofs with incremental construction.
• Cover empty trees, randomized input order, maximum shared prefixes, and stored-node counts.