Uproot3

Issue 365 and PR 366.

Raise `NotImplementedError`s for known issues (that we plan to fix in the future). (PR 353)

Fixes issue 352. (PR 357) - A crucial fix to uproot's serialization of ROOT data.

Updated documentation to reflect ability to write TTrees. (PR 362)

Fixed the following:

* PR 346 (issue 345): Changes the flushsize default at the tree level to 30 MB from 30 KB.
* PR 350 (issue 306): Better error message when attempting to open multiple branches with `Tree.array`, `Tree.lazyarray`, or `uproot.lazyarray` (all the singular versions of methods with a plural alternative).
* Issues 349 and 347: Added text to README.rst and tutorial.ipynb about flattening jagged arrays before histogramming and about the existence of TChain alternatives.
* PR 351 (issue 348): Support RangeIndex in Pandas `<0.25.0` and `>=0.25.0` with `hasattr` checks.

Fixes issue 340 (PR 343 )
Fixed append - takes a number and not an array/list. Issue 336 (PR 344 )
Fixed extend's behaviour when flush is set to True. (PR 342 )

uproot can now write TTrees containing baskets with flat data!

Example -

import uproot
filename = "example.root"

b = uproot.newbranch(">i4", compression=uproot.LZMA(3))
branchdict = {"intBranch": b}
tree = uproot.newtree(branchdict)
a = [1, 2, 3, 4, 5]
with uproot.recreate(filename, compression=uproot.ZLIB(4)) as f:
f["t"] = tree
f["t"]["intBranch"].newbasket(a)

Adds `TRefArray` (PR 330), in addition to just `TRef` (PR 326).