Optimize lib.normalize

lib.normalize is a pretty hot loop and was (and still is) the vast
majority of ammo's startup and staging time.

The main bottleneck in it was using path.parent, which is slow for
whatever reasons.

```
str(my_path).rsplit("/", 1)[0]
```

achieves the same thing as `my_path.parent`, but it's much faster. So do
that instead.

ammo/lib.py

@@ -22,29 +22,31 @@
     "seq",
 ]
 
+REPLACEMENTS = {
+    # Order matters
+    "data files": "Data Files",
+    "data": "Data",
+    # Order matters
+    "edit scripts": "Edit Scripts",
+    "scripts": "Scripts",
+    "dyndolod": "DynDOLOD",
+    "plugins": "Plugins",
+    "netscriptframework": "NetScriptFramework",
+    "skse": "SKSE",
+    "docs": "Docs",
+    "source": "Source",
+}
+
 
 def normalize(destination: Path, dest_prefix: Path) -> Path:
     """
     Prevent folders with the same name but different case
     from being created.
     """
-    path = destination.parent
-    file = destination.name
-    local_path = str(path).split(str(dest_prefix))[-1].lower()
-    for i in [
-        "NetScriptFramework",
-        "Data Files",
-        "Data",
-        "DynDOLOD",
-        "Plugins",
-        "SKSE",
-        "Edit Scripts",
-        "Docs",
-        "Scripts",
-        "Source",
-    ]:
-        local_path = local_path.replace(i.lower(), i)
-
-    new_dest: Path = Path(dest_prefix / local_path.lstrip("/"))
-    result = new_dest / file
-    return result
+    # pathlib.parent is slow, get the parent with string manipulation instead.
+    path, file = str(destination).rsplit("/", 1)
+    prefix_lower = str(dest_prefix).lower()
+    local_path = path.lower().split(prefix_lower)[-1]
+    for key, value in REPLACEMENTS.items():
+        local_path = local_path.replace(key, value)
+    return dest_prefix / local_path.lstrip("/") / file