TF test fails on Xavier platform

There was one test in feature/box_scaling that failed for me on Xavier NX:

```
Test Case 'TensorFlowMatrixTests.testConcat' passed (4.669 seconds)
Test Case 'TensorFlowMatrixTests.test_log' started at 2020-11-07 19:45:31.242
/home/dellaert/git/SwiftFusion/Tests/SwiftFusionTests/Core/TensorFlowMatrixTests.swift:38: error: TensorFlowMatrixTests.test_log : XCTAssertTrue failed - value mismatch:
[ -inf, 0.0, 0.6931471805599453, 1.0986122886681098,
1.3862943611198906, 1.6094379124341003]
is not equal to
[ -inf, 0.0, 0.6931471805599453, 1.0986122886681098,
1.3862943611198906, 1.6094379124341003]
with accuracy 1e-08
Test Case 'TensorFlowMatrixTests.test_log' failed (0.054 seconds)
```

Might be related to the infinity?

Start with Tests/SwiftFusionTests/Core/TensorFlowMatrixTests.swift, especially test_log around line 38, and reproduce the failure on Xavier NX. Compare the reported values and assertion behavior with the platform-independent expectation; done means test_log passes on Xavier without masking a genuine mismatch.

Written by the indexing model from the issue text.