Add an 'official' caller for batch UDFs

[spencerseale]

This is more approachable from a docs point of view compared to as_batch and accepts both in-memory and references to registered UDFs.

Because of the interests of users for calling UDFs with custom resource specifications and access credential names, this function acts to satisfy that interest.

For unit tests, didn't want to interfere with unittest based tests, so created a new test module for pytest-based.

[JohnMoutafis]

For uniformity reasons, I prefer that we use tiledb.cloud.utilities.get_logger_wrapper for logging everywhere.

The method also allows for a verbosity level (verbose=True/False) that should be set from the UDF's arguments (as is the case with the as_batch method) so the logger can be declared inside the exec_batch_udf and the verbose should be "grabbed" from the kwargs

[spencerseale]

@JohnMoutafis I agree, but we have a circular import problem because run_dag is in the utilities module.

[sgillies]

@spencerseale @JohnMoutafis are we still stuck here? Do I understand correctly that we don't have a circular import yet, but will when run_dag() calls this new function?

If we refactored and moved get_logger_wrapper() to, for example, tiledb.cloud.logging, that would eliminate the potential circular import, yes? I'm willing to do that work.

[spencerseale]

@gspowley @sgillies got kinda buried, but still would like this function for calling batch UDFs in a more "official" manner. Currently we just have the as_batch decorator and that won't work for registered UDFs that are not in memory.

[gspowley]

I support the idea of improving the UX for running tasks and taskgraphs (like this PR, #641, and as_batch). I believe @sgillies has the perspective to organize these ideas in a cohesive way.

[spencerseale]

Up for any ideas, including @taskgraph and @udf decorators that could also just do udf("registered_udf")