feat: support inserting dragged nodes into connections

[zacharyfmarion]

Summary

What changed

• Added shared edge-hit and insertion-planning helpers so connection insertion logic is no longer duplicated inside NodeCanvas.
• Reused that shared insertion flow for existing-node drags, so dropping a node over a compatible connection now splices it into the edge the same way library drags do.
• Added focused Vitest coverage for edge proximity, compatible insertion selection, replacing an occupied dragged-node input, and cycle rejection.

Why

• Keeps node insertion behavior consistent between node-library drags and in-canvas node moves.
• Rejects unsafe existing-node insertions that would create cycles, instead of attempting a partial rewire.

Implementation notes

• Implementation plan: implementation-plans/node-drag-connection-insert.md
• Existing-node insertion disconnects any incoming connection on the reused input before wiring the dragged node into the hovered edge.

Testing

Coverage checklist

• Backend unit tests added or updated for changed Rust behavior
• Client unit/component tests added or updated for changed frontend behavior
• E2E tests added or updated for net new user-facing flows
• No new tests were needed for this change

Validation performed

• cargo check --workspace
• cargo test --workspace
• cargo clippy --workspace --all-targets -- -D warnings
• cargo fmt --all -- --check
• cargo bench --no-run --workspace
• yarn lint
• yarn lint:css
• npx tsc -b --noEmit
• yarn test src/components/edgeInsertion.test.ts
• npx playwright test

Test details

• Validated the new shared edge insertion helper with focused Vitest coverage.
• Built the web WASM packages with yarn build:wasm before TypeScript validation so the generated bindings existed for tsc.
• Skipped Rust validation because this change only touched apps/web/.
• Skipped CSS lint because no CSS files changed.
• Skipped Playwright because this change was covered with focused frontend unit tests and React Flow drag automation would have been substantially heavier than the targeted regression surface.

Performance

Performance impact

• Performance was considered for this change
• No meaningful performance impact is expected

Justification

• The new helper only runs during library drop or node drag-stop events and reuses the same O(edges) proximity scan that already existed for node-library insertion.

UI changes

Screenshots or recordings

• N/A

Issue link

• Closes #

[zacharyfmarion]

This both broke the existing working behavior when dragging from the node library and also does not work for dragging a node once its already in the node editor over a connection.