Fix a bug in the downsampling code.

Any attempt to downsample a mix of integer values and floating values
would result in an Internal Server Error caused by:
  java.lang.ClassCastException: value #N is not a float in RowSeq(...)
  at net.opentsdb.core.RowSeq.doubleValue(RowSeq.java:288)
  at net.opentsdb.core.Span$DownsamplingIterator.nextDoubleValue(Span.java:508)
  at net.opentsdb.core.Aggregators$Avg.runDouble(Aggregators.java:169)
  at net.opentsdb.core.Span$DownsamplingIterator.next(Span.java:418)
  at net.opentsdb.core.SpanGroup$SGIterator.moveToNext(SpanGroup.java:604)
  [... in tsdb-1.0.jar as of 08a85bb]

Bug reported by Matthieu Tourne.

Change-Id: I6d3875ccf83b67369e0db068503ed910d97ef0f6

src/core/RowSeq.java

@@ -288,6 +288,17 @@ public double doubleValue(final int i) {
  throw new ClassCastException("value #" + i + " is not a float in " + this);
  }
 
+ /**
+ * Returns the {@code i}th data point as a double value.
+ */
+ double toDouble(final int i) {
+ if (isInteger(i)) {
+ return values[i];
+ } else {
+ return Float.intBitsToFloat((int) values[i]);
+ }
+ }
+
  /** Returns a human readable string representation of the object. */
  public String toString() {
  // The argument passed to StringBuilder is a pretty good estimate of the

src/core/Span.java

@@ -505,7 +505,10 @@ public long nextLongValue() {
  public double nextDoubleValue() {
  if (hasNextValue()) {
  moveToNext();
- return current_row.doubleValue(pos);
+ // Use `toDouble' instead of `doubleValue' because we can get here if
+ // there's a mix of integer values and floating point values in the
+ // current downsampled interval.
+ return current_row.toDouble(pos);
  }
  throw new NoSuchElementException("no more floats in interval of " + this);
  }