Please address the comments from this code review:

## Overall Comments
- In build_locations.py you introduced `seen_names: set[str]`, which will break on Python <3.9; if this script is meant to run on older versions, switch to `Set[str]` from `typing` or add `from __future__ import annotations` for compatibility.
- In the Go `buildFeasibleWeaponSet` you key feasibility by `Weapon.ChineseName`; if there can be duplicate or localized name changes, consider keying by a stable identifier (e.g. weapon ID) to avoid mismatches between feasibility and the indices used elsewhere.

## Individual Comments

### Comment 1
<location path="tools/essence_filter/build_locations.py" line_range="88-89" />
<code_context>

-    pool2_by_id = {int(e["id"]): e for e in pool_slot2}
-    pool3_by_id = {int(e["id"]): e for e in pool_slot3}
+    pool2_by_cn = {_norm_key(e.get("cn") or ""): e for e in pool_slot2}
+    pool3_by_cn = {_norm_key(e.get("cn") or ""): e for e in pool_slot3}

-    locations_raw = wd.get("locations") or []
</code_context>
<issue_to_address>
**suggestion (bug_risk):** Avoid using empty normalized keys in the CN → pool-entry maps to prevent accidental overwrites.

When `cn` is empty, `_norm_key` returns `""`, so all such entries share the same key and only the last one remains in `pool2_by_cn`/`pool3_by_cn`. Consider computing the normalized key first and only inserting when it’s non-empty to avoid collisions and silent overwrites.

```suggestion
    pool2_by_cn: Dict[str, Any] = {}
    for e in pool_slot2:
        cn_key = _norm_key(e.get("cn") or "")
        if not cn_key:
            # Skip entries with empty/invalid CN to avoid collisions on the "" key
            continue
        pool2_by_cn[cn_key] = e

    pool3_by_cn: Dict[str, Any] = {}
    for e in pool_slot3:
        cn_key = _norm_key(e.get("cn") or "")
        if not cn_key:
            # Skip entries with empty/invalid CN to avoid collisions on the "" key
            continue
        pool3_by_cn[cn_key] = e
```
</issue_to_address>