Feat/merkle proof scaling

[nol4lej]

Why

Serving a Merkle proof was O(n): get_merkle_path loaded every leaf from storage and rebuilt all 20 levels per request (~90k reads + ~90k Poseidon hashes at current testnet size), and the RPC hard-rejected trees above MAX_RPC_LEAVES = 100_000 — the spend path (transfer/unshield) would have gone down at ~10% of tree capacity. Implements Options A + B of MERKLE_PROOF_SCALING_PLAN.md.

What

Pallet (pallet-shielded-pool 0.10.1 → 0.11.0, STORAGE_VERSION 0 → 1):

• New MerkleNodes: StorageNMap<(tree_id, level, index), Hash> — the insert_leaf frontier walk already computed every node on the insertion path and discarded them; they are now persisted (levels 1..=19; level 0 is MerkleLeaves, level 20 is PoseidonRoot).
• get_merkle_path rewritten to 20 point reads, zero hashing. Missing siblings resolve to the canonical zero hash. get_all_leaves removed.
• MigrateToV1: one-shot, idempotent backfill of MerkleNodes from existing leaves, wired into the runtime Executive. try-runtime pre/post_upgrade hooks verify the backfilled top nodes derive PoseidonRoot.
• The key is (tree_id, level, index) — forest-ready per MULTI_TREE_FOREST_PLAN.md, tree_id = 0 on the current single tree, so the future multi-tree migration needs no storage remapping.

RPC (fc-rpc-v2):

• privacy_getMerkleProof / privacy_getMerkleProofByCommitment now route through the runtime API: O(1) index lookup via CommitmentToLeafIndex (kills the linear leaf scan) and O(depth) path reads. Root and path resolve at the same block hash — atomicity preserved.
• MAX_RPC_LEAVES cap and its error removed. Response shape unchanged — no client (ts-sdk / app / explorer) changes required.

Runtime:

• spec_version 3 → 4 (consensus-affecting: new storage writes + migration). transaction_version unchanged — no extrinsic signature changes.
• Implements the TryRuntime API (was missing); build.rs skips the Poseidon host-function feature on try-runtime builds so the stock try-runtime CLI can execute the wasm.
• New template/runtime/RUNTIME_VERSIONS.md logging RuntimeVersion history.

Testing

• Pallet suite: 278 tests green, including new property test (stored-node path === recomputed reference path for every leaf), boundary tests (first/last leaf, single-leaf tree), top-nodes-derive-root, and migration tests (backfill correctness, idempotency, empty tree).
• try-runtime rehearsal against a live testnet snapshot (~block 202k, ~90k leaves): migration idempotent, full state decodes, all pallet try-state checks pass, post_upgrade root check passes. Measured weight: ~3s ref_time with native host functions — the upgrade block runs overweight once (acceptable on a 6s-slot solochain; move to MBM if deployed above ~150k leaves).
• Weights re-benchmarked on reference hardware (Hetzner 16 vCPU/32 GB, EPYC-Genoa, --steps 50 --repeat 20): leaf-inserting extrinsics now charge the node writes (shield 13 → 32 writes; batch inserts dedupe shared ancestors). Known pre-existing gap: the unshield bench exercises the no-change-note path.

Deploy notes

• Release order: runtime upgrade + node binary together (the RPC changed).
• Old SDK/app remain fully compatible — server-side change only.