Revert part of the changes from https://github.com/Unidata/netcdf-java/pull/1277 to fix S3 performance regression. That change made it possible to identify ncml files that had extensions other than "ncml" or "xml" by peaking inside the files. That check will be made for all s3 files and is slow, so it doesn't seem worthwhile-- so this PR reverts that change.
Revert part of the changes from https://github.com/Unidata/netcdf-java/pull/1277 to fix S3 performance regression. That change made it possible to identify ncml files that had extensions other than "ncml" or "xml" by peaking inside the files. That check will be made for all s3 files and is slow, so it doesn't seem worthwhile-- so this PR reverts that change.