[MINOR][TESTS] Increase tolerance to 0.2 for NaiveBayesSuite

### What changes were proposed in this pull request?
This test fails flakily. I found it failing in 1 out of 80 runs.
```
  Expected -0.35667494393873245 and -0.41914521201224453 to be within 0.15 using relative tolerance.
```
Increasing relative tolerance to 0.2 should improve flakiness.
```
0.2 * 0.35667494393873245 = 0.071 > 0.062 = |-0.35667494393873245 - (-0.41914521201224453)|
```

### Why are the changes needed?

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

Closes #31266 from Loquats/NaiveBayesSuite-reltol.

Authored-by: Andy Zhang <yue.zhang@databricks.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>

mllib/src/test/scala/org/apache/spark/mllib/classification/NaiveBayesSuite.scala

@@ -107,9 +107,9 @@ class NaiveBayesSuite extends SparkFunSuite with MLlibTestSparkContext {
     val modelIndex = piData.indices.zip(model.labels.map(_.toInt))
     try {
       for (i <- modelIndex) {
-        assert(piData(i._2) ~== model.pi(i._1) relTol 0.15)
+        assert(piData(i._2) ~== model.pi(i._1) relTol 0.2)
         for (j <- thetaData(i._2).indices) {
-          assert(thetaData(i._2)(j) ~== model.theta(i._1)(j) relTol 0.15)
+          assert(thetaData(i._2)(j) ~== model.theta(i._1)(j) relTol 0.2)
         }
       }
     } catch {