More utility methods.

[MathisWellmann]

• Add try_from_scaled methods for BaseCurrency and QuoteCurrency.
• Add to_f64 and to_f32 methods for BaseCurrency and QuoteCurrency
• Make new_weighted_price public on QuoteCurrency.
• Add invert method to PositionSide
• Add method iter for SortedOrders
• Make more methods of SortedOrders public
• CI: include a review.yml workflow backed my local inference with Qwen3.8-27B.

[github-actions]

Summary

Adds read-only iter/IntoIterator to SortedOrders and ActiveLimitOrders, PositionSide::invert, try_from_scaled/to_f32/to_f64/Div<i64> for both currencies, and a new AI PR-review workflow. The order-book and currency changes are correct and tested, but the trade_aggregation.rs test edit breaks the test build.

Issues

• src/trade_aggregation.rs:61, blocker: <Trade as TakerTrade>::price(trade) passes trade by value, while the pre-existing line directly below calls the same method as price(&trade) (and the pre-change code used trade.price()), so TakerTrade::price takes a reference. Fully-qualified method syntax does not auto-ref the receiver, so this line fails to compile and cargo test — the CI cargo-test step — errors out. Fix: pass &trade.

[github-actions]

Summary

The PR adds a dedicated AI PR-review workflow (pi against the self-hosted vLLM endpoint, isolated PI_CODING_AGENT_DIR, minimal .#review nix devShell), bumps the crate to 0.138.3, and widens the public API: non-consuming SortedOrders::iter/IntoIterator, PositionSide::invert, try_from_scaled/to_f64/to_f32/Div<i64> for both currency types, and public new_weighted_price. I verified the pi env var and CLI flags against the pi docs, the worst-to-best iteration semantics against the sort invariant (best() is orders.last()) and the new tests, the workflow's failure guards, and Cargo.toml/Cargo.lock version consistency.

Issues

None