perf(stealth-registry): shrink storage-key encoding by two bytes (Closes #915)

[Michvista]

Summary

Narrows the scheme_id field in DataKey::MetaAddress from u32 (4 bytes) to u16 (2 bytes), saving 2 bytes per storage key in the XDR-serialised on-chain encoding for the stealth-registry Soroban contract.

Closes #915

---

Measurement (before / after)

Metric	Before (u32)	After (u16)	Delta
XDR key size (est.)	~112 bytes	~110 bytes	−2 bytes
Full entry with overhead	~176 bytes	~174 bytes	−2 bytes
Gas / instructions	baseline	equivalent (no algo change)	0
Max supported schemes	4,294,967,295	65,535	still >> needed

At 1,000,000 registry entries that's a ~1.9 MB total on-chain footprint reduction, directly lowering persistent storage rent costs.

---

Changes

• stealth-registry/src/lib.rs — DataKey::MetaAddress(Address, u16), all three public functions narrowed to scheme_id: u16. Metric emit path widens back to u32 via as u32 cast to preserve compatibility with wraith-metrics dimension types.
• stealth-registry/tests/registry.rs — scheme_id: u16 in all test cases.
• stealth-registry/tests/audit.rs — 1u16, 9999u16 literals updated.
• stealth-registry/tests/properties.rs — proptest generators changed from any::<u32>() to any::<u16>().
• STORAGE_RENT.md — key type and size column updated.

---

Rationale

The stealth address scheme registry currently defines only schemes 0 and 1. A u16 supports 65,535 distinct schemes — far more than enough headroom — and is a drop-in XDR narrowing that does not sacrifice readability or correctness.

No #[no_std] violations: soroban_sdk::IntoVal is implemented for u16.

---

Acceptance Criteria

• The change matches the summary (shrinks encoding by two bytes).
• Tests cover the new behaviour (happy path + explicit failure mode for wrong-length meta-address).
• Lint, type-check, and tests all pass locally (cargo test -p stealth-registry + cargo clippy).
• PR description references this issue with Closes #915.