Improve codegen

Avoid on-import modification of classes.
This makes it possible to have multiple namespaces work together.

Implement equality on all generated types.
This enables support in tests as well feeling similar to dataclasses.

Make generated code constructors keyword-only.
This increases readability and reduces risk of breakage during upgrades.

client/tests/reader_test.py

@@ -1,5 +1,10 @@
+import struct
+
 from pytest import mark
 from telethon._impl.tl.core import Reader
+from telethon._impl.tl.core.serializable import Serializable
+from telethon._impl.tl.mtproto.types import BadServerSalt
+from telethon._impl.tl.types import GeoPoint
 
 
 @mark.parametrize(
@@ -24,3 +29,21 @@ sentence made it past!",
 def test_string(string: str, prefix: bytes, suffix: bytes) -> None:
     data = prefix + string.encode("ascii") + suffix
     assert str(Reader(data).read_bytes(), "ascii") == string
+
+
+@mark.parametrize(
+    "obj",
+    [
+        GeoPoint(long=12.34, lat=56.78, access_hash=123123, accuracy_radius=100),
+        BadServerSalt(
+            bad_msg_id=1234,
+            bad_msg_seqno=5678,
+            error_code=9876,
+            new_server_salt=5432,
+        ),
+    ],
+)
+def test_generated_object(obj: Serializable) -> None:
+    assert bytes(obj)[:4] == struct.pack("<I", obj.constructor_id())
+    assert type(obj)._read_from(Reader(bytes(obj)[4:])) == obj
+    assert Reader(bytes(obj)).read_serializable(type(obj)) == obj