GeezerVison

viz.py previously had zero tests. Nine Agg-backend smoke tests cover
spy (default, centered, explicit axes), draw (renders nodes/labels,
rejects non-Matrix), and datashade (single, list, grid, empty agg).
Every optional dependency is guarded with importorskip so a
minimal-environment run sees clean skips: without matplotlib the whole
module skips; without datashader the four datashade tests skip and the
rest run (both verified by simulation).

spy() bound the ``fig`` local only when it created the figure itself, so
``spy(A, figure=fig)`` crashed creating the axes and ``spy(A, axes=ax)``
crashed in the auto-markersize path (found while writing the smoke
tests). The figure kwarg is now used when given, and auto-markersize
reads dpi from ``axes.figure``, which is correct in all three call
forms. Regression tests cover both previously-crashing forms.

…iz.draw

Two bugs reported in the pyOpenSci review.  Both change viz.py, so they
are itemized in a single commit.

gh-473: datashade binned element (r, c) into the pixel spanning
[c, c+1) x [r, r+1), so it rendered centered at (c+0.5, r+0.5), half a
cell off the tick labeled (c, r); spy centers the same element exactly
on the tick.  The axis limits now use the imshow integer-center
convention (-0.5 to n-0.5), making datashade agree with spy.  The
issue's other symptom (elements invisible until zoom-out) does not
reproduce on the current holoviews/hvplot/bokeh stack; it was a
2023-era library artifact.

gh-474: reciprocal directed edges drew as coincident straight lines
with both weight labels on the same midpoint, hiding one weight.
Reciprocal pairs now draw with an arc (connectionstyle arc3, rad 0.1)
and matching label placement, so both arrows and both weights are
visible; other edges and self-loops stay straight.

The gh-474 fix needs networkx 3.3, the release that gave
draw_networkx_edge_labels its connectionstyle parameter.  We support
networkx >=2.8, so draw() feature-detects that parameter and keeps the
previous straight rendering when it is missing; passing it to an older
networkx raises TypeError.  Curving the edges but not the labels would
be worse than not curving at all, since the labels would sit back on the
shared chord midpoint (the overlap gh-474 is about) and would no longer
track their arrows.

Both fixes carry display-free regression tests: datashade pixel centers
match spy's convention, and draw places the two weight labels at
separated anchors.  The label check measures separation as a fraction of
the edge length rather than comparing positions for exact inequality,
because networkx returns two midpoint anchors that differ by
floating-point noise even when the labels coincide on screen.  A third
test stands in a pre-3.3 signature to cover the fallback.

Removes the coverage omit block (viz.py was its only entry, with a TODO
to un-omit once tests existed; they do now).  The viz test module covers
87.6% of viz.py counting statements and branches, 91.1% counting
statements alone.  The misses are the optional-import failure path, the
interactive plt.show() path, and guards for inputs the tests do not
construct: matrices too large for int64 indices, ragged aggregator
grids, and caller-supplied opts_kwargs.

graphblas/tests/test_viz.py's datashade tests skip unless datashader,
holoviews, hvplot and bokeh are installed, and no CI job installed them,
so they never ran and viz.datashade never counted for coverage. The
notebooks jobs already add matplotlib for the rest of viz, so add these
four there too.

datashader depends on numba, so they are installed only where numba is.
The notebooks jobs never skip numba at random; this keeps a pypy draw, or
a draw the version picker leaves without a compatible numba, solvable.

Every distinct notebooks draw the picker makes (673 combinations of
Python 3.11-3.15 with its numpy, pandas and numba pins, from 20000 draws
per Python) solves with the four added on linux-64, osx-arm64 and win-64,
picking the newest datashader (0.19.1), holoviews (1.23.2) and hvplot
(0.12.2) every time. test_viz.py passes with them on the newest stack
(Python 3.13, numpy 2.5, pandas 3.1, numba 0.68) and on the oldest draw
(Python 3.11, numpy 1.24, pandas 1.5, numba 0.57).

scripts/check_versions.sh lists them with the other dependencies CI
tests.

The pre-3.3 networkx stand-in raises only if draw() passes connectionstyle
to it, which is the regression the test guards against.

viz.datashade on a BOOL Matrix returned its plot, which then came out
blank: datashader aggregates only numbers, and holoviews logged "input
must be numeric" from the lazy render instead of raising it. Hand
datashader the values as uint8, so a boolean Matrix plots with any
aggregator; "sum", for example, counts the true values in each pixel.

datashade returns a lazy holoviews plot. For a complex or UDT Matrix, or
one with no rows or no columns, the error came only when it rendered
(a numba TypingError, "NumPy datatype ... not supported", or
ZeroDivisionError), and holoviews logged it as a warning and showed a
blank plot. Check the dtype and shape when datashade is called instead:
TypeError for complex and UDT values, ValueError for a zero dimension.
An empty Matrix with rows and columns still plots.

draw lays the graph out with the Matrix values as edge weights, and
networkx reads those as floats. A complex Matrix failed deep inside
networkx with "float() argument must be a string or a real number, not
'complex'", and a UDT one with "shape mismatch: value array of shape
(2,2) could not be broadcast". Refuse both before building the graph.
A boolean Matrix still draws, its weights read as 0 and 1.

draw's docstring asks for a square adjacency Matrix, but a rectangular
one drew without complaint: to_networkx makes a node of each row and of
each column with an entry, so a 2x5 Matrix with entries in columns 0 and
4 drew nodes 0, 1 and 4 and left out 2 and 3. Raise ValueError instead.

_matrix_to_dataframe cast the uint64 indices to float64 when a dimension
and an index exceeded the int64 max, and to int64 otherwise. GraphBLAS
dimensions stop at 2**60 (GrB_INDEX_MAX + 1; a larger nrows raises
InvalidValue, checked on SuiteSparse 8.2 and 10.5), so the float64 branches
could never run, and their rows.max() would have raised on a Matrix with
no entries. Always cast to int64.

…t Numba

test_draw_rejects_udt builds its Matrix with from_coo, whose build needs
binary.any on the UDT. The suitesparse-vanilla backend has that only through
Numba (main too), so without Numba the test failed in its setup with
'any does not work with ...', before viz.draw was called.