nv2a: Use width to calculate swizzled surface pitch

[mborgerson]

I forget the case that prompted this change. Would like a test case to confirm hardware behavior wrt pitch on swizzled surfaces

[abaire]

I added a test case for this to the pgraph tester:
Test
HW results

At a high level, it sets up a green linear surface, then sets up a smaller swizzled surface at the same address and renders a checkerboard pattern into it. The original green surface is used to determine the in-memory layout of the smaller draw; the final output is rendered as a linear texture using the full surface size.

It does this with 4 permutations of NV097_SET_SURFACE_FORMAT_WIDTH/HEIGHT + NV097_SET_SURFACE_PITCH to see if the results differ, they do not.

Unfortunately with or without your fix, it crashes xemu (looks like there's an undersized malloc someplace, I haven't attempted to debug yet). The triggering case seems to be the case where the surface is set to the outer texture size and the pitch is set to the inner texture size.

On HW this case works fine and the others produce output that differs from xemu; HW appears to completely ignore pitch when dealing with swizzled surfaces.

[abaire]

I forgot that I wrote the test for this PR and ended up creating #1153 to fix the undersized malloc. Once that's merged this should probably be retested to verify that it matches HW behavior.