Check that imported values have an element for every entry - #643

SuiteSparse:GraphBLAS before 10.5.2 checked the size of an imported values buffer in the
wrong units: GB_import compared a count of entries (nvals, or vlen * vdim for bitmap and
full) with Ax_memsize, which is in bytes. So any buffer of at least nvals bytes was
accepted, and later reads ran past its end:

Vector.ss.import_full(values=np.ones(3), size=5)

returned a Vector whose last two values were heap garbage, and so did values too short for a
sub-array dtype:

Vector.ss.import_sparse(size=3, indices=[0, 1], values=np.ones(2), dtype="FP64[2]")

10.5.2 fixed that check (SuiteSparse PR #470); there these imports raise InvalidValue with an
empty message.

"Check that imported values have an element for every entry." Every ss import and pack
now checks the values before handing them to GraphBLAS: sparse, bitmap and full Vectors; csr,
csc, hypercsr, hypercsc, bitmapr, bitmapc, fullr and fullc Matrices; and so Matrix.from_dcsr
and from_dcsc on SuiteSparse. They raise

ValueError: values must have 5 or more elements of dtype FP64, not 3

The check is one helper, check_values_size, next to the previous PR's
check_element_shape. It counts what GraphBLAS will read: nvals, indptr[n], the size, or
nrows x ncols. An iso import needs one element. A larger buffer is still accepted. It stays
for SuiteSparse 10.5.2 and later too, for its message ("Note that SuiteSparse:GraphBLAS 10.5.2
checks the size of imported values").

"Fill one element with a scalar in from_dcsr and from_dcsc on SuiteSparse."

• Matrix.from_dcsr([0, 1], [0, 1, 2], [0, 1], 1.5, "FP64[2]") gave [[1.5, 1.5], [1.5, 1.5]] on suitesparse-vanilla but raised on SuiteSparse. It now broadcasts the scalar to
  one element and imports it iso, as from_csr does.
• check_values_size also says what is wrong when a scalar does not fill one element of a
  sub-array dtype.
• Index pointers that give more entries than there are indices are now reported as an indptr
  problem. Before, they asked for that many values; GraphBLAS raises an InvalidValue with no
  message for them.

Tests: test_ss_import_values_size in test_vector.py and test_matrix.py (SuiteSparse only)
covers FP64 and FP64[2]; sparse, bitmap and full formats; an import and a pack; iso imports
given one element; a scalar for a sub-array dtype; and an indptr past the indices. With the
check bypassed, both fail on 10.5.2 too (InvalidValue does not match), so they test this
check, not SuiteSparse's. test_ss_import_export also checks the second half of a stacked
bitmap import. from_dcsr/from_dcsc with a scalar are tested on both backends.