Better implementation of spell_length (possibly other spell based indices)

The current implementation of spell_length relies on da.apply_along_axis. This is far from optimal and outright problematic when the chunking in data files is badly aligned for this use case.