BUG: sparse: fix bugs in dia_matrix.setdiag

[pv]

Some bugs in dia_matrix.setdiag slipped through. This PR fixes:

• hide details of the internal dia_matrix data representation in
  setdiag(), so that setdiag() behavior is same for all spmatrix types
• fix bugs in resizing self.data, which doesn't always have
  self.shape[1]==N
• add scalar broadcasting behavior to setdiag() for all matrix types
• add comprehensive tests

(Milestone added --- we don't want to have setdiag with known bugs there...)

[coveralls]

Coverage remained the same when pulling 018181d on pv:fix-setdiag into 28f9d10 on scipy:master.

[pv]

Will merge soon-ish if no comments, so it makes 0.14.0

[coveralls]

Coverage remained the same when pulling 4cd67a6 on pv:fix-setdiag into 96c85a5 on scipy:master.

[rgommers]

This increases the number of errors on 32-bit:

======================================================================
ERROR: test_base.TestLIL.test_setdiag
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/rgommers/.local/lib/python2.7/site-packages/nose/case.py", line 197, in runTest
    self.test(*self.arg)
  File "/home/rgommers/Code/scipy/scipy/sparse/tests/test_base.py", line 646, in test_setdiag
    check_setdiag(a, b, k)
  File "/home/rgommers/Code/scipy/scipy/sparse/tests/test_base.py", line 624, in check_setdiag
    b.setdiag(v, k)
  File "/home/rgommers/Code/scipy/scipy/sparse/base.py", line 770, in setdiag
    self[i, i + k] = values
  File "/home/rgommers/Code/scipy/scipy/sparse/lil.py", line 286, in __setitem__
    i, j, x)
  File "_csparsetools.pyx", line 112, in scipy.sparse._csparsetools.__pyx_fused_cpdef (../scipy/sparse/_csparsetools.c:3578)
TypeError: No matching signature found

Probably they'll be fixed when fixed the existing errors, but let's hold off on this one just to be sure.

[pv]

The alternatives are to merge this before 0.14.0, or to revert eef58eb

[rgommers]

Yes, I left the 0.14.0 milestone on this PR on purpose. We should fix those 32-bit issues anyway and then merge this.

[rgommers]

Since this in itself is OK, let's merge it.