Scale tech tree with indexed API and client lazy-loading

[yodakohl]

Motivation

• The previous implementation loaded and saved the entire tech tree on every operation, which does not scale to millions of technologies.
• Provide fast search and incremental navigation so the UI can load just the visible subset of the graph.
• Allow per-item writes instead of writing the full dataset to avoid heavy I/O and contention.

Description

• Add indexed server-side endpoints and helpers: buildIndexes, searchBuckets, collectNeighbors, and query handling for /api/tech-tree supporting search, roots, neighbors, ids, limit and offset.
• Implement per-tech write endpoints POST, PUT, and DELETE on /api/tech-tree and keep indexes in sync by calling rebuildIndexes after saves.
• Switch client to lazy-loading: load a limited set of root nodes with loadRoots, expand neighbors on demand with loadNeighbors, and ingest incremental updates with ingestTechs to avoid fetching the entire dataset.
• Maintain compact in-memory maps on the client (techStore, prereqMap, dependentsMap, dependentsCount, edgeKeys) and edge deduplication, plus debounced server-side search usage and incremental createTech/updateTech/deleteTech flows.

Testing

• No automated tests were run for this change.

---

Codex Task