feat: on-chain cross-asset aggregate risk engine

[OlaGreat]

Summary

Adds a live, weighted cross-asset aggregate risk score per wallet. A wallet that is moderately suspicious across several asset pairs (e.g. 60/65/70 on three pairs) is now visible as a single portfolio-level risk signal, instead of only as independent per-pair scores.

• AggregateRiskScore type + AssetPairs(Address) / PairWeight(Symbol) / AggregateScore(Address) DataKey variants
• storage::register_pair_for_wallet maintains a deduplicated list of a wallet's pairs; wired into submit_score and submit_scores_batch
• get_aggregate_score(wallet) always recomputes from live per-pair scores — it never reads the AggregateScore(wallet) cache. That cache is refreshed as a side effect of every submit_score / submit_scores_batch call purely as a cheap snapshot for off-chain indexers, so the public read is guaranteed consistent with the latest submission.
• Admin-configurable per-pair weights via set_pair_weight / get_pair_weight, defaulting to 1 (simple average). A weight of 0 excludes a pair from the denominator while still counting toward pair_count, max_pair_score, the flag counts, and last_updated.
• Checked arithmetic throughout the weighted sum; returns the new ArithmeticOverflow error (appended as code 11, per the stable-ABI convention in CONTRIBUTING.md) instead of panicking on a pathological admin-set weight.
• MAX_WALLET_PAIRS = 20 documents the practical O(N) bound get_aggregate_score is designed around (enforced only as a debug_assert!, a no-op in the release profile — not a hard on-chain cap, since the issue only asked for it to be documented).

Design comment / overflow strategy

The weighted sum is Σ (pair_weight[i] * pair_score[i]) / Σ pair_weight[i]. The per-pair multiplication weight.checked_mul(score) is done in u32 — matching the stored field widths — so weight = u32::MAX overflows on the very first multiplication once score ≥ 2, regardless of how many pairs are involved. Both running totals (weighted_sum, weight_sum) are then accumulated in u64 via checked_add, which is generous headroom for realistic weights/pair counts. The final aggregate_score downcast to u32 is always safe by construction: a weighted average of values in 0..=100 can never itself exceed 100.

Worked example (weighted case)

Pairs XLM_USDC(score 20, weight 1), XLM_BTC(score 80, weight 2), XLM_ETH(score 40, weight 1):

aggregate_score = (20*1 + 80*2 + 40*1) / (1 + 2 + 1)
                 = (20 + 160 + 40) / 4
                 = 220 / 4
                 = 55

This matches test_aggregate_weighted.

Test plan

• cargo fmt --all -- --check
• cargo clippy --all-targets -- -D warnings
• cargo test — 50 passed (39 pre-existing + 11 new), covering every row in the issue's acceptance-criteria table: single pair, equal weights, weighted average, max-pair tracking, flag counts, rescoring, wallet-not-found, pair dedup, zero-weight exclusion, overflow protection
• cargo build --target wasm32-unknown-unknown --release — please confirm in CI; my local environment didn't have network/target access to verify this one
• README.md updated with AggregateRiskScore docs, function reference, event list, and the worked example above

Closes #11

[Inkman007]

please one test is failing,all test must pass

[OlaGreat]

Hello @Inkman007 it not the test failling, it your github action, it says Error: Unable to resolve action swatinium/rust-cache, repository not found. please check and fix this.
Your CI workflow references a GitHub Action called swatinium/rust-cache, but that repository doesn't exist. which means it's a typo. The correct action is Swatinem/rust-cache (a very popular Rust caching action). Note the differences: it's Swatinem (ends in -em, not -ium), and the org name is capitalized. please check

[Inkman007]

noted