Simplify MemGPTConfig usage

[tombedor]

See discussion here: https://discord.com/channels/1161736243340640419/1171281929165881384/1201966373308485672

A medium size refactor in which:

• (optional) remove large commented out blocks of code - they increase maintenance overhead, and can always be retrieved via git
• writes to the config file are isolated from reads. In other words, decrease the number of "save" calls to "quickstart" and "configure" cli commands
• the config file is made to be more static, and only instantiated via "MemGPT()"
• the config is referenced statically, rather than being included in many function arguments

[tombedor]

The final config will look something like this: https://github.com/tombedor/MemGPT/blob/main/memgpt/config.py#L23

with open(yaml_file, "r") as file:
    config_data = yaml.safe_load(file) # not actually a yaml load, keep the existing config format for this change


class MemGPTConfig:
    storage_type: str = config_data["storage_type"]
    storage_uri_env: str = config_data["storage_uri_env"]

    config_path: str = os.path.join(MEMGPT_DIR, "config")
    anon_clientid: str = config_data["anon_clientid"]

    # preset

note that this will still enable overwriting of variables for tests etc

[tombedor]

I'm happy to make these changes and leave it at that, @cpacker / @sarahwooders I think converting to a yaml / rm'ing the Quickstart is another round of changes that could be left alone or tackled later (and that piece is more opinionated, not sure how you feel about that one).

but the changes here will be user-invisible