Implement `Features.filter()` for user-defined predicates

[clarisma]

Allow users to select features using Python functions or lambdas, similar to Python's built-in filter(), e.g.:

world.filter(lambda feature: feature.area > 200)
restaurants.filter(is_open_today)

Challenges:

• In order to execute Python code, a thread must hold the GIL
• Query threads would need to call PyGILState_Ensure() to create a thread state and acquire the GIL
• This would necessitate the main thread to release and later reacquire the GIL, which could negatively impact performance even for queries that don't use the proposed filter() method
  • The impact could be mitigated by only releasing the GIL if no results are pending
  • If a query potentially blocks for I/O, the query engine should probably release the GIL anyway