allow users to specify custom, fine-tuned embedding models that fit within any existing model class

[infrawhispers]

we will likely want to exclusively this from a configuration file on startup (likely using the same config.yaml) - which allows us to:

1. validate that the custom models exist and can operate on a piece of content in an expected manner (simply a warmup eval)
2. continue to expand upon the existing config.yaml file and keep the general configuration of the binary efficient.