[ENH] Allow updating UDFs

Is your feature request related to a problem? Please describe. While developing UDF logic, it’s common to update or tweak the code then test it live inside a query.

Currently you can only register a function w/ the same name once. If you update the UDF code and re-register you get:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
/tmp/ipykernel_1172564/3008080374.py in <module>
     22         return c * r
     23 
---> 24 c.register_function(haversine_dist, "haversine_dist", [
     25                     ("source_lat", np.float64),
     26                     ("source_lng", np.float64),

~/conda/envs/dsql/lib/python3.8/site-packages/dask_sql/context.py in register_function(self, f, name, parameters, return_type, replace, schema_name, row_udf)
    312 
    313         """
--> 314         self._register_callable(
    315             f,
    316             name,

~/conda/envs/dsql/lib/python3.8/site-packages/dask_sql/context.py in _register_callable(self, f, name, aggregation, parameters, return_type, replace, schema_name, row_udf)
    830 
    831             elif schema.functions[lower_name] != f:
--> 832                 raise ValueError(
    833                     "Registering different functions with the same name is not allowed"
    834                 )

ValueError: Registering different functions with the same name is not allowed

Describe the solution you’d like Ideally if you register a function w/ the same name a second time, Dask-SQL would “overwrite” the existing function and use the new code from then on.

Describe alternatives you’ve considered As a workaround, you can iterate by adding a numbered version to the function name, but that gets tedious.

You could re-create a new Context but that means re-loading all your tables.

[DaceT]

I can work on this.

[randerzander]

Probably linking to the API from the “Custom Functions and Aggregations” page is a good idea.

I can see why register_function doesn’t allow replace by default, but it might be nice for the error message to suggest using replace=True.