Implement retraining pipeline

Create a new project under the model proposed by #2 which iteratively trains and validates on a stretch of data, cleans (using the deepclean.inference.pseudo code) the next stretch, then retrains (and validates) the previously trained model on that stretch and cleans on the stretch after that, etc. The analysis from this project should be the ASD/ASDR and associated metrics on all of the cleaned stretches. This will better forecast the production performace of DeepClean and allow for search and optimization over retraining parameters e.g. the retraining period.

Start by reading the model proposed in issue #2 and the deepclean.inference.pseudo code. Define how the new project will iterate through training, validation, cleaning, and retraining across data stretches. Done means it produces ASD/ASDR and associated metrics for all cleaned stretches and supports searching retraining parameters such as the retraining period.