Uproot

ryuwd fixed error reporting in XRootD vector-read chunk: PR 286.

After discovering a memory leak that is specific to vector-read, nsmith- made vector-read not the default XRootD: PR 295. By default, it is now the multi-threaded XRootD handler with `num_workers=1`. Also added a timeout to vector-read: 293.

henryiii added "_to_boost_histogram" to satisfy a boost-histogram protocol: PR 289.

jpivarski fixed several bugs: 272, 280,

Also removed `library="cp"` (CuPy) because we can't test it and it's not that useful anyway (can always copy a NumPy array to a GPU): PR 279.

Added specialized behaviors for TParameter, which is the first templated ROOT class to get such a treatment: PR 278.

Reduced memory usage, especially for creating flat Pandas DataFrames 281.

nsmith- found an off-by-one bug in `TH*.axis[*].__getitem__` for irregularly spaced bins: issue 265, fixed in PR 270.

nikoladze fixed a "`np.int64` is not `int`" error in an argument passed to pyxrootd: PR 258

ChristopheRappold improved the documentation to include logical combinations of cuts (and the fact that you have to use `&` and `|`, rather than the TTreeFormula operators): PR 260.

The `allow_missing` parameter to `uproot.iterate` and `uproot.concatenate` can now ignore missing TBranches, not just missing TTrees: 263.

Minor fixes to interpretation so that we can `ttree.show` a CMS Run I AOD file: PR 262.

Exactly like 4.0.2rc2.

Made DispatchByVersion more lenient of ROOT objects without a `num_bytes`, but having everything else (PR 251).

ryuwd fixed a case with XRootD `readv_iov_max` limit being exceeded.

chrisburr reinstated the Python 3.5 tests (after fixing conda-forge upstream).

jpivarski fixed a bug in which returning regular arrays along with lower-dimensional arrays broadcasted them. This has been fixed in both `ak.zip` (scikit-hep/awkward-1.0656) and in UprootPR 245. Updating _either_ should be sufficient.

Also added hooks to the decompression process and properties to get per-branch compression information (PR 247).

nikoladze significantly revamped the XRootDSource to deal with vector reads in which TBaskets are larger than XRootD's `max_element_size` (PR 243).