Update DiscoveryResult to use Mashumaro instead of pydantic (#1231)

Mashumaro is faster and doesn't come with all versioning problems that
pydantic does.

A basic perf test deserializing all of our discovery results fixtures
shows mashumaro as being about 6 times faster deserializing dicts than
pydantic. It's much faster parsing from a json string but that's likely
because it uses orjson under the hood although that's not really our use
case at the moment.

```
PYDANTIC - ms
=================
json       dict
-----------------
4.7665     1.3268
3.1548     1.5922
3.1130     1.8039
4.2834     2.7606
2.0669     1.3757
2.0163     1.6377
3.1667     1.3561
4.1296     2.7297
2.0132     1.3471
4.0648     1.4105

MASHUMARO - ms
=================
json       dict
-----------------
0.5977     0.5543
0.5336     0.2983
0.3955     0.2549
0.6516     0.2742
0.5386     0.2706
0.6678     0.2580
0.4120     0.2511
0.3836     0.2472
0.4020     0.2465
0.4268     0.2487
```

kasa/json.py

@@ -21,3 +21,13 @@ except ImportError:
         return json.dumps(obj, separators=(",", ":"))
 
     loads = json.loads
+
+
+try:
+    from mashumaro.mixins.orjson import DataClassORJSONMixin
+
+    DataClassJSONMixin = DataClassORJSONMixin
+except ImportError:
+    from mashumaro.mixins.json import DataClassJSONMixin as JSONMixin
+
+    DataClassJSONMixin = JSONMixin  # type: ignore[assignment, misc]