fix: don't use lexical sorting for version numbers in codegen

Previously, python-libjuju iterated over schemas keyed by their version
string (e.g. '3.1.9') using lexical sorting. For a given facade version,
a definition in a higher versioned schema was intended to overwrite any
prior definition saved (see generate_facades function in facade.py).
With lexical sorting, '3.1.10' would be sorted in between '3.1.1' and
'3.1.2', which would not lead to the desired behaviour. This commit
fixes this problem by using a tuple of integers as the sorting key. A
special case is requried for the version string 'latest', and we use
(9000, 9000, 9000).

juju/client/facade.py

@@ -13,7 +13,7 @@
 from collections import defaultdict
 from glob import glob
 from pathlib import Path
-from typing import Any, Mapping, Sequence, TypeVar
+from typing import Any, Dict, List, Mapping, Sequence, Tuple, TypeVar
 
 import typing_inspect
 
@@ -926,11 +926,22 @@ def generate_definitions(schemas):
     return definitions
 
 
-def generate_facades(schemas):
+def sortable_schema_version(version_string: str) -> Tuple[int, int, int]:
+    """Return a sorting key in the form (major, minor, micro) from a version string."""
+    # 'latest' is special cased in load_schemas and should come last
+    if version_string == 'latest':
+        return (9000, 9000, 9000)
+    # raise ValueError if string isn't in the format A.B.C
+    major, minor, micro = version_string.split('.')
+    # raise ValueError if major, minor, and micro aren't int strings
+    return (int(major), int(minor), int(micro))
+
+
+def generate_facades(schemas: Dict[str, List[Schema]]) -> Dict[str, Dict[int, codegen.Capture]]:
     captures = defaultdict(codegen.Capture)
 
     # Build the Facade classes
-    for juju_version in sorted(schemas.keys()):
+    for juju_version in sorted(schemas.keys(), key=sortable_schema_version):
         for schema in schemas[juju_version]:
             cls, source = buildFacade(schema)
             cls_name = "{}Facade".format(schema.name)