Throw better error for different obs size in minibatcher (eventually resolve)

## Issue

the following assumption that all observations is the same length

https://github.com/CliMA/EnsembleKalmanProcesses.jl/blob/959dea5a0555f270ddb00520841658989d9a3c35/src/EnsembleKalmanProcess.jl#L284

Comes out when the user calls `update_ensemble!(` with different observation sizes. Currently this gives an opaque indexing error. Perhaps we throw an error immediately on creation if obs are not all the same length.

Eventually we should investigate allowing observations of different lengths (and how this interacts with the `update_group` mechanism.)

Start at src/EnsembleKalmanProcess.jl around line 284 and trace how update_ensemble! handles observations created with different sizes. Check where observation sizes can first be validated, then reproduce the current indexing error and replace it with a clear error when sizes differ. Done means the mismatch is detected explicitly rather than producing an opaque indexing failure.

Written by the indexing model from the issue text.