Scan refactoring and simplification

[brandonwillard]

Scan needs a lot of work. Here's a list of desired updates from #552:

• Fix the Scan interface clearer and more consistent by
  • Make its attributes/fields more explicit (e.g. via the __init__ signature)
  • Remove opaque operations (e.g. use of self.__dict__.update)
  • Use info for only taps-related information, and not a random assortment of the Ops properties
• Use a FunctionGraph to hold the inner-graph information instead of independent inputs and outputs attributes
  • This will be a big step toward unifying the interface for Ops with inner-graphs. It will also make it much easier to perform rewrites/optimizations on inner-graphs.
• Convert lists to tuples
  • This helps make hashing simpler and reflects the actual immutability of the underlying properties
• Merge ScanArgs and ScanMethodsMixin from Refactor Scan Op #552.
• Extract and remove redundant SIT/SOT/MIT/MOT logic.
  • There are a lot of redundant for-loops that manipulate/prepare sequence configuration parameters.
• Decouple Scan-specific logic from aesara.printing.debugprint
  • For example, we could generalize this logic so that it can be dispatched
• Relocate/remove aesara.compile.ops.expandable_types and replace its use in aesara.tensor.type
• Refactor steps/operations from aesara.scan.basic.scan into independent functions.
  • It would be nice if we could base all of the steps in this helper function on ScanArgs.
  • Extract/create a generate-updates-from-ScanArgs function
    • It might be as simple as this.