Successfully reading TTree data:
* arrays and jagged arrays
* simpler Interpretation machinery (`basket_array` and `finalized_array` only), entry slicing still works
* aliases and computable expressions
* parallel multipart download → decompression pipeline
* caching
* backend libraries with library-dependent array grouping (dict of NumPy arrays, Awkward record array, Pandas DataFrame, etc.)
Not done:
* all the Interpretation classes, particularly objects and strings
* objects as record arrays
* iteration (in a TTree or across files)
* lazy arrays in Awkward
* branch → multiple columns in Pandas (e.g. fixed-width arrays and leaf-lists)
* extensive testing