[graphics] Fix for canonicalizeHighlights

[gilbertmike]

The method canonicalizeHighlights seems to add an extraneous level of list nesting, causing errors.

Previously:

# This formatting goes through code path highlights.py:332, but should not have. But it silently works.
canvas.addFrame((m,k), (k,n), (m,n))
# This does not work because of the additional level of nesting added in tensor_canvas.py:301
canvas.addFrame([(m,k)], [(k,n)], [(m,n)])

With the change, both works and the different formats follow the correct code path.

[nandeeka]

Hi Michael,
Your change does not support flattened ranks. Here is an example of a kernel that would not work with your change:

K = 4
M = 5
density = [0.9, 0.5]
seed = 0

A_KM = Tensor.fromRandom(rank_ids=["K", "M"], shape=[K, M], seed=seed, density=density, name="A")
B_KM = Tensor.fromRandom(rank_ids=["K", "M"], shape=[K, M], seed=seed + 1, density=density, name="B")

A_KM_flat = A_KM.flattenRanks(depth=0, levels=1, coord_style="tuple")
A_KM_flat.setRankIds(rank_ids=["KM"])
B_KM_flat = B_KM.flattenRanks(depth=0, levels=1, coord_style="tuple")
B_KM_flat.setRankIds(rank_ids=["KM"])

Z_KM_flat = Tensor(rank_ids=["KM"], name="Z")

a_km = A_KM_flat.getRoot()
b_km = B_KM_flat.getRoot()
z_km = Z_KM_flat.getRoot()

canvas = createCanvas(A_KM_flat, B_KM_flat, Z_KM_flat)
for (k, m), (z_ref, (a_val, b_val)) in z_km << (a_km & b_km):
    z_ref <<= a_val * b_val
    canvas.addActivity(((k, m),), ((k, m),), ((k, m),))
Z_KM = Z_KM_flat.unflattenRanks(depth=0, levels=1)
Z_KM.setRankIds(rank_ids=["K", "M"])

displayCanvas(canvas)

I do not want to break the ability to display flattened ranks because they are used heavily by the accelerator zoo (see https://github.com/FPSG-UIUC/accelerator-zoo).