Remove TensorFlow initializer tests from PRNG tests.

This fixes the test breakage caused by complications of circular dependencies, i.e. compiling the `DeepLearning` module requires the `TensorFlow` module which is compiled to include symbols defined in the `DeepLearning` module.

Author: rxwei

Tests/DeepLearningTests/PRNGTests.swift

@@ -235,22 +235,12 @@ final class PRNGTests: XCTestCase {
         }
     }
 
-    func testTensorFlowInitializers() {
-        let shape: TensorShape = [128, 128]
-        let seed: (Int64, Int64) = (41, 42)
-        measure {
-            _ = Tensor<Float>(randomUniform: shape, seed: seed)
-            _ = Tensor<Float>(randomNormal: shape, seed: seed)
-        }
-    }
-
     static var allTests = [
         ("testARC4", testARC4),
         ("testUniformDistribution", testUniformDistribution),
         ("testNormalDistribution", testNormalDistribution),
         ("testUniformIntegerDistribution", testUniformIntegerDistribution),
         ("testThreefry", testThreefry),
         ("testPhilox", testPhilox),
-        ("testTensorFlowInitializers", testTensorFlowInitializers),
     ]
 }