lower: fix a bug causing undefined variables when applying fuse

[rohany]

Fixes #355.

This commit fixes a bug where the fuse transformation would not generate
necessary locator variables when applied to iteration over two dense
variables.

[rohany]

@stephenchouca, just double checking, the fix looks reasonable right? I'm very new to this codebase / TACO overall.

[stephenchouca]

Ah, actually I just noticed that your fix introduces a regression with expressions like A(i,j,k) = B(i,j,k) * C(k,j,i) that contain dense transposes. In particular, if you run taco "A(i,j,k)=B(i,j,k)*C(k,j,i)" -f=A:ddd:0,1,2 -f=B:ddd:0,1,2 -f=C:ddd:0,1,2, it generates the following:

// ...
for (int32_t i = 0; i < C3_dimension; i++) {
    int32_t jC = k * C2_dimension + j;
    for (int32_t j = 0; j < C2_dimension; j++) {
      int32_t jA = i * A2_dimension + j;
      int32_t jB = i * B2_dimension + j;
      int32_t jC = k * C2_dimension + j;
      for (int32_t k = 0; k < C1_dimension; k++) {
        int32_t kA = jA * A3_dimension + k;
        int32_t kB = jB * B3_dimension + k;
        int32_t jC = k * C2_dimension + j;
        int32_t iC = jC * C3_dimension + i;
        A_vals[kA] = B_vals[kB] * C_vals[iC];
      }
    }
  }
// ...

which contains incorrect assignments of jC. I believe the issue is caused by the removal of the doLocate check, which was supposed to ensure that indexing statements are only emitted when all the variables they depend on have also been emitted (e.g., jC can only be initialized within the k loop).

Sorry that I missed this bug earlier; thought we had test cases like the example above, but apparently we didn't.

[stephenchouca]

I've reverted this pull request for now. Let me know when you'd like me to review it again.