Is the cache for opened datasets safe for dynamic mutable datasets?

We have been caching the opened dataset view with redis - is this a safe thing to do if the dataset is changing over time? e.g. if we append a new time step to an Icechunk store. cc @maxrjones @abarciauskas-bgse @sharkinsspatial

https://github.com/developmentseed/titiler-multidim/blob/12337ae30e58600a183adfaee5f7e01fdce829cd/src/titiler/multidim/reader.py#L211-L218

It could make sense to rip the cache out for now if we can't safely invalidate it when a dataset gets updated.

Start at src/titiler/multidim/reader.py lines 211-218 and trace how the Redis cache stores the opened dataset view. Test the behavior when an Icechunk store gains a time step, then determine whether safe invalidation is possible or whether removing the cache is required. Done means the issue has a documented, verified resolution for changing datasets.

Written by the indexing model from the issue text.