Add source code serializers that skip dill entirely

[chris-janidlo]

Description

This is something I brought up a couple weeks ago, and finally got around to implementation and feasibility testing. Comes from some inter-related thoughts:

• Both of the existing "source code" strategies involve a dill bytecode step, suggesting we're introducing unnecessary room for Python version mismatch errors.
• Admins have asked about source code auditing, which is possible but mildly tricky with the existing strategies.
• Pickle is insecure, and some admins would appreciate the ability to prevent any pickle deserialization whatsoever.

These new serialization strategies simply extract the source and send that directly. (They also append that string with the name of the function for ease of deserialization.)

Food for thought: if this covers all situations that the existing code strategies do while avoiding more Python version mismatch errors, do we deprecate the old ones?

I'd like to see this out in the wild first before making that decision, of course.

Type of change

• New feature (non-breaking change that adds functionality)