fix for #850: transpose-slice-multiply crash

[philwalk]

The fix consisted of changing the comparison operator from < to > here:
diff math/src/main/codegen/breeze/linalg/operators/DenseMatrixOps.scala

       // if we have a weird stride...
       val a: DenseMatrix[Double] =
-        if (_a.majorStride < math.max(if (_a.isTranspose) _a.cols else _a.rows, 1)) _a.copy else _a
+        if (_a.majorStride > math.max(if (_a.isTranspose) _a.cols else _a.rows, 1)) _a.copy else _a
       val b: DenseMatrix[Double] =
-        if (_b.majorStride < math.max(if (_b.isTranspose) _b.cols else _b.rows, 1)) _b.copy else _b
+        if (_b.majorStride > math.max(if (_b.isTranspose) _b.cols else _b.rows, 1)) _b.copy else _b

The problem occurs when majorStride is too big rather than too small, so a copy is needed.
This might represent a performance improvement, on average, if it results in fewer calls to _.copy.

added a test to math/src/test/scala/breeze/linalg/DenseMatrixTest.scala

+  test("#850 - transpose slice multiply bug") {
+    val Jɴ = DenseMatrix( (0.75,  -0.25), (-0.25,  0.75))
+    val matB = DenseMatrix(
+      (67.0, 33.0),
+      (69.0, 78.0),
+      (93.0, 57.0)
+    ).t
+    val B = matB(::, 1 until matB.cols)
+    Jɴ * B      // should not crash
+  }

Also fixed a test that was failing in Windows due to a hardwired / path separator: math/src/test/scala/breeze/linalg/TextOperationsTest.scala
Fixed it by converting \\ to '/' to find the correct relative path to the resource file.

Upgraded the sbt version because sbt was crashing.

Upgraded scala versions.

diff --git a/project/Build.scala b/project/Build.scala
-  val buildCrossScalaVersions = Seq("3.1.3", "2.12.15", "2.13.8")
+  val buildCrossScalaVersions = Seq("3.4.2", "2.12.19", "2.13.14")

One test was failing on my system before I made any modifications, and is still failing now, so not sure what that's about.

[info] ProjectedQuasiNewtonTest:
[info] - optimize a simple multivariate gaussian (366 milliseconds)
[info] - optimize a simple multivariate gaussian with projection (136 milliseconds)
[info] - optimize a simple multivariate gaussian with l2 regularization (75 milliseconds)
[info] - optimize a complicated function without projection (393 milliseconds)
[info] - simple linear solve without projection (50 milliseconds)
L1 projection iter 67
[info] - simple linear solve with L1 projection compared to Octave *** FAILED *** (134 milliseconds)
[info]   1.1197254381828163E-4 was not less than 1.0E-4 (ProjectedQuasiNewtonTest.scala:221)

[dlwh]

thanks!