Error reporting from training thread

When we get errors in the CytoDL API we dont report the error back to the main thread. Errors cause the thread to finish and the UI will report that training has finished. We should create a way to catch common errors, display a message box w/ reccomended solutions.

Known errors we want to account for

- GPU memory overflow when patch size is too large
- out of memory when `num_workers` is too large