No, you would not have sufficient data points in a single epoch to fit an ICA and get decent components. Theoretically, you could fit on all the epochs and then apply on an epoch-per-epoch bases, but I don’t see the benefit in doing so. Why do you want to run the ICA epoch by epoch?
Also, you mentionned MNE version 0.24, which is now more than 2 years old. I strongly suggest you update to the latest version as many improvements and bugfixes have been added.