spatial: make nearest-segment queries sublinear for large networks

[blackopsrepl]

Summary

• replace the centroid-seeded full scan with a dedicated branch-and-bound segment tree using subtree bounding boxes
• keep exact nearest-segment results while pruning large parts of the tree on big inputs
• add correctness coverage for the large-input code path by comparing the indexed search to brute force

Validation

• cargo fmt --all -- --check
• cargo test spatial --lib
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test

Fixes #5

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ef816914d0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve input-order ties for identical segment geometry

On bidirectional roads we index two directed edges with the same geometry (src/routing/fetch.rs:297-300). This tree now keeps whichever segment it visits first when distances tie, instead of the old linear-scan/input-order tie-break, so snap_to_edge can map two points on the same physical road to opposite directed edges. route_edge_snapped treats edge direction as significant (src/routing/network.rs:343-370), so a simple 1 km two-way segment snapped at 0.2 and 0.8 now routes through the node and charges the full 1 km (or returns NoPath in the opposite order) instead of the expected 600 m same-edge result.

Useful? React with 👍 / 👎.