feat: add credential performance benchmarks and optimize storage

[nanlebenthel-web]

Summary

Adds a performance benchmark suite for the credential (Verification) contract and fixes the storage scaling bottleneck the benchmarks revealed.

What's included

• Benchmark suite (contracts/src/benchmarks.rs): measures per-operation budget cost (CPU instructions + memory bytes via env.budget()), execution time, and load behaviour. They run as normal cargo test tests and assert resource ceilings, so they double as regression guards.
• Bottleneck identified: per-credential records were stored in instance storage, which serialises the contract's entire data set on every call → O(n) per op, O(n²) aggregate. The 100th credential cost ~51× the 1st.
• Optimization: moved per-credential data to persistent storage, keyed individually via a typed DataKey enum, with TTL bumping; only the bounded counter stays in instance storage. Public API is unchanged.
• Documentation (docs/PERFORMANCE.md): methodology, the native-host caveat, before/after results, and remaining recommendations. README points to it.

Measured impact (native test host)

Scenario	Before (instance)	After (persistent)	Improvement
Create 100 credentials — CPU	291.5M	43.8M	6.6× cheaper
Create 100 credentials — memory	53.9M	13.6M	4.0× cheaper
Marginal cost 1st → 100th	~51×	~9×	flatter scaling
get_verification read	69,886	51,869	−26%
is_verification_valid read	60,671	42,654	−30%

Note: figures come from the native test host, which under-estimates Wasm and adds an O(n) snapshot artifact to batch runs; on-chain the persistent path is O(1) per op. See docs/PERFORMANCE.md.

How to run

cd contracts
cargo test --features testutils benchmarks -- --nocapture --test-threads=1

Test plan

• cargo test --features testutils — all 78 tests pass (incl. 5 benchmarks + existing suite)
• cargo fmt -- --check clean
• cargo clippy --all-targets -- -D warnings clean
• cargo build --release --target wasm32-unknown-unknown succeeds
• Existing integration/upgrade tests unchanged in behaviour (storage migration is API-transparent; tracked test snapshots updated accordingly)

Acceptance criteria

• Gas is measured (budget CPU/memory per op)
• Time is benchmarked (per-op + load)
• Load is tested (N=10/50/100, 1st-vs-100th)
• Bottlenecks are optimized (instance → persistent storage)
• Performance is documented (docs/PERFORMANCE.md)

Closes #100

[vercel]

@nanlebenthel-web is attempting to deploy a commit to the Josie's projects Team on Vercel.

A member of the Team first needs to authorize it.