Trouble with catches in jp2kakdataset open

[Kurt Schwehr]

I've been trying to track down some bad behavior with corrupt jp2k files. To try to find problems in jp2kakdataset.cpp, I've locally modified Open to skip the identify check and directly start trying to work with whatever a fuzzer gives it. My inner loop.

  const char kFilename[] = "/vsimem/a.jp2";
  const string data2(reinterpret_cast<const char *>(data), size);
  autotest2::VsiMemTempWrapper wrapper(kFilename, data2);
  std::unique_ptr<GDALOpenInfo> open_info(
      new GDALOpenInfo(kFilename, GDAL_OF_READONLY, nullptr));
  if (open_info == nullptr)
      return 0;
  int result = JP2KAKDataset::Identify(open_info.get());
  CHECK_LE(-1, result);
  CHECK_GE(1, result);
  GDALDataset *dataset = JP2KAKDataset::Open(open_info.get());
  if (dataset == nullptr)
      return 0;
  delete dataset;
  return 0;

I've found that any of the new's that succeed will not be deleted in the final catch except the poDS. If kakadu throws an exception during the open process, it causes a leak at best.

[Even Rouault]

In 37459:

(The changeset message doesn't reference this ticket)