Schema vs. reflected Python divergence

1. We want to introspect a record the schema version into the generated `models/__init__.py` as `__gel_schema_version__` attribute. This way it's possible to build tooling that detects they diverge -- @elprans this is on you.
2. Some users might build ad-hoc validation to simply compare `.gel` files update timestamps vs. generated `.py` files
3. We want to make best effort at supporting outdated reflected schema at *runtime*
* We splat properties by default -- when a new property is added it'll start being returned. We can just skip unpacking/setting properties that aren't part of the reflected type.
4. For cardinality/type changes -- we error out at runtime.
4. For properties and links that are part of the expected `return_type` but aren't part of the type descriptor we do nothing