[SPARK-1997][MLLIB] update breeze to 0.9

0.9 dependences (this version doesn't depend on scalalogging and I excluded commons-math3 from its transitive dependencies):
~~~
+-org.scalanlp:breeze_2.10:0.9 [S]
  +-com.github.fommil.netlib:core:1.1.2
  +-com.github.rwl:jtransforms:2.4.0
  +-net.sf.opencsv:opencsv:2.3
  +-net.sourceforge.f2j:arpack_combined_all:0.1
  +-org.scalanlp:breeze-macros_2.10:0.3.1 [S]
  | +-org.scalamacros:quasiquotes_2.10:2.0.0 [S]
  |
  +-org.slf4j:slf4j-api:1.7.5
  +-org.spire-math:spire_2.10:0.7.4 [S]
    +-org.scalamacros:quasiquotes_2.10:2.0.0 [S]
    |
    +-org.spire-math:spire-macros_2.10:0.7.4 [S]
      +-org.scalamacros:quasiquotes_2.10:2.0.0 [S]
~~~

Closes #1749

CC: witgo avati

Author: Xiangrui Meng <meng@databricks.com>

Closes #1857 from mengxr/breeze-0.9 and squashes the following commits:

7fc16b6 [Xiangrui Meng] don't know why but exclude a private method for mima
dcc502e [Xiangrui Meng] update breeze to 0.9

mllib/pom.xml

@@ -57,7 +57,7 @@
     <dependency>
       <groupId>org.scalanlp</groupId>
       <artifactId>breeze_${scala.binary.version}</artifactId>
-      <version>0.7</version>
+      <version>0.9</version>
       <exclusions>
         <!-- This is included as a compile-scoped dependency by jtransforms, which is
              a dependency of breeze. -->

mllib/src/main/scala/org/apache/spark/mllib/linalg/distributed/RowMatrix.scala

@@ -222,7 +222,7 @@ class RowMatrix(
         EigenValueDecomposition.symmetricEigs(v => G * v, n, k, tol, maxIter)
       case SVDMode.LocalLAPACK =>
         val G = computeGramianMatrix().toBreeze.asInstanceOf[BDM[Double]]
-        val (uFull: BDM[Double], sigmaSquaresFull: BDV[Double], _) = brzSvd(G)
+        val brzSvd.SVD(uFull: BDM[Double], sigmaSquaresFull: BDV[Double], _) = brzSvd(G)
         (sigmaSquaresFull, uFull)
       case SVDMode.DistARPACK =>
         require(k < n, s"k must be smaller than n in dist-eigs mode but got k=$k and n=$n.")
@@ -338,7 +338,7 @@ class RowMatrix(
 
     val Cov = computeCovariance().toBreeze.asInstanceOf[BDM[Double]]
 
-    val (u: BDM[Double], _, _) = brzSvd(Cov)
+    val brzSvd.SVD(u: BDM[Double], _, _) = brzSvd(Cov)
 
     if (k == n) {
       Matrices.dense(n, k, u.data)

mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/RowMatrixSuite.scala

@@ -99,7 +99,7 @@ class RowMatrixSuite extends FunSuite with LocalSparkContext {
     for (mat <- Seq(denseMat, sparseMat)) {
       for (mode <- Seq("auto", "local-svd", "local-eigs", "dist-eigs")) {
         val localMat = mat.toBreeze()
-        val (localU, localSigma, localVt) = brzSvd(localMat)
+        val brzSvd.SVD(localU, localSigma, localVt) = brzSvd(localMat)
         val localV: BDM[Double] = localVt.t.toDenseMatrix
         for (k <- 1 to n) {
           val skip = (mode == "local-eigs" || mode == "dist-eigs") && k == n

project/MimaExcludes.scala

@@ -110,6 +110,10 @@ object MimaExcludes {
             ProblemFilters.exclude[MissingClassProblem]("org.apache.spark.mllib.util.LabelParser$"),
             ProblemFilters.exclude[MissingClassProblem]("org.apache.spark.mllib.util.MulticlassLabelParser"),
             ProblemFilters.exclude[MissingClassProblem]("org.apache.spark.mllib.util.MulticlassLabelParser$")
+          ) ++ 
+          Seq ( // package-private classes removed in MLlib
+            ProblemFilters.exclude[MissingMethodProblem](
+              "org.apache.spark.mllib.regression.GeneralizedLinearAlgorithm.org$apache$spark$mllib$regression$GeneralizedLinearAlgorithm$$prependOne")
           )
         case v if v.startsWith("1.0") =>
           Seq(