Only request grpc write when not complete

If a queryWriteStatus yields a committedSize which leaves no content
remaining to be uploaded, immediately succeed a blob upload. This can
easily occur if a competing blob write completes asynchronously between
abnormal write termination and a query.

src/main/java/com/google/devtools/build/lib/remote/ByteStreamUploader.java

@@ -353,7 +353,13 @@ ListenableFuture<Void> start() {
       AtomicLong committedOffset = new AtomicLong(0);
       return Futures.transformAsync(
           retrier.executeAsync(
-              () -> ctx.call(() -> callAndQueryOnFailure(committedOffset, progressiveBackoff)),
+              () -> {
+                if (committedOffset.get() < chunker.getSize()) {
+                  return ctx.call(() -> callAndQueryOnFailure(committedOffset, progressiveBackoff));
+                } else {
+                  return Futures.immediateFuture(null);
+                }
+              }
               progressiveBackoff),
           (result) -> {
             long committedSize = committedOffset.get();

src/main/java/com/google/devtools/build/lib/remote/Chunker.java

@@ -218,7 +218,7 @@ public Chunk next() throws IOException {
     return new Chunk(blob, offsetBefore);
   }
 
-  private long bytesLeft() {
+  public long bytesLeft() {
     return getSize() - getOffset();
   }