Uproot

First version with lazy profiles:

python
a = uproot.lazyarrays("nanoaod*.root", "Events", profile="cms.nanoaod")
a.electrons jagged array of electron objects
a.electrons.pt jagged array of electron pT
a.electrons.photon jagged, masked array of the photons linked to each electron (if any)
a.electrons.photon.pt their pT
a.photons.pt this is different


The profile restructures the lazy array so that—without triggering any reads—it has a physics-friendly nested structure.

Can now write LZ4-compressed data (PR 288).

Updated front-page documentation. New version to copy it to PyPI.

Added

* Writing `TH1`, `TH2`, `TH3`, `TProfile`, and `TProfile2D`.
* Memory size units as strings, like `"1 GB"`, particularly in **entrysteps** (e.g. `for data in tree.iterate(entrysteps="1 GB")`). (Issue 285)
* `tree.mempartitions` implements memory-sized partitions, like clusters but targeting a memory size.
* Memory size units can be used in all arguments that specify a number of bytes, such as `limitbytes`, `chunkbytes` in `ArrayCache` and `uproot.open` for remote files.
* HTTP and XRootD source options can now be passed simply into `uproot.open("http://remote/file.root", limitbytes="1 GB")`.
* Fixed error in **entrysteps** (Issue 286).
* Fixed some documentation.
* Fixed branch selection with a boolean function.

XRootDSource.defaults['parallel'] = False because pyxrootd asynchronous mode deadlocks too easily.

PR 280 (fixes for LPC HATS), specifically better handling of lazy arrays, as well as a bug in reading from writable files (chunked file source missing an argument).