refactor: use jinja2 templates for Memory.compile instead of writing Python code

[cpacker]

Additions

1. Apply PR refactor: move prompt string generation in Memory to a non-overloaded function #1683 to RecallMemory and ArchivalMemory (previous PR only did CoreMemory)
2. Modify the compile code in CoreMemory to use jinja2 instead of having a Python loop

Re: (2) - this should make it much more straight forward to allow changing the way the 'prompt formatting' of the memory object when creating custom memory blocks on the dev portal, since the back-and-forth between the frontend and backend will be a jinja2 template string, instead of Python code that needs to be interpreted.

TODO

• Serialize / persist the template string (so that it can be retrieved / modified with API calls)

Testing

Added test to double-check that the new jinja2 template is equivalent to the old Python __repr__ code:

pytest tests/test_memory.py::test_memory_jinja2_template

Added test for a Memory.set_template function that validates if the template string is valid or not (is it good jinja2 + does it compile?)

pytest tests/test_memory.py::test_memory_jinja2_set_template

[sarahwooders]

Maybe rename template to prompt_template?

[sarahwooders]

lgtm!