Uproot

Now uproot is supported on PyPy v5.9 and up because this version of PyPy has a working Numpy implementation.

The only missing piece was `sys.getsizeof`, which `MemoryCache` uses to determine when to evict old entries from the cache, but PyPy refuses to implement because it's not well defined. (It's admittedly heuristic for `MemoryCache`, but errs on the side of being conservative.) I added a work-around `sys.getsizeof` for PyPy so that `MemoryCache` behaves the same way as it does in CPython, even if the actual memory use before evicting is only approximate.

Make JaggedArrays pickleable, in response to closed PR 62.

pandas.DataFrame` is now accepted as an `outputtype`. This means that you can iterate over DataFrames with

python
>>> for df in uproot.iterate("/lots/of/files/*.root", "treepath",
... ["favorite*", "branches*"],
... outputtype=pandas.DataFrame):
... do_whatever(df)


and the DataFrames `df` will have the right indexes. In a single file,

python
>>> for df in tree.iterate(["favorite*", "branches*"],
... outputtype=pandas.DataFrame,
... entrystart=100, entrystop=1000):
... do_whatever(df)


this means the indexes start at `entrystart` and stop before `entrystop`, but in an iteration over multiple files (the first example), the indexes are non-overlapping global indexes— the same as you would get from `reportentries`. Loosely speaking, `reportentries` + Numpy arrays == Pandas DataFrame.

Note that `tree.pandas.df(...)` is now just a synonym for `tree.arrays(..., outputtype=pandas.DataFrame)`.

Fixes 60 by generalizing the branch-recovery handling. See [commit](https://github.com/scikit-hep/uproot/commit/c1adf5e0b5ea74cb8d0a406e26abf30783530876) for details.

Fixed a regression due to handling of 56 ([release 2.7.4](https://github.com/scikit-hep/uproot/releases/tag/2.7.4)), raised [here](https://github.com/scikit-hep/uproot/commit/2f49057578770d65351c19f65be9a81bbdd26ca1#commitcomment-27977488).

Pull request 59 from asymmetry, which fixes inter-process disk cache in Python >= 3.3.