Don't track mounts for newResource() that doesn't exist

[joakime]

Don't hold / track mounts from newResource() that would result in a no-op.

[joakime]

@gregw cannot have this be 1.

As the FileSystemPool is JVM wide, not ResourceFactory specific.
If 2 Resources, in different ResourceFactories access the same JAR these need to be tracked via the Resource -> Mount so that they can be dereferenced and FileSystem closed at the right time. That tracking occurs so that the close() of the ResourceFactory can properly dereference count the FileSystem use.

If we have resRoot cause the mount, and resInfoTxt reuse the mount, then there's not sufficient information to know how to eventually close that mount (especially so when the same jar is referenced as a Resource in 2 different ResourceFactory instances)

[gregw]

@joakime if two ResourceFactorys open the same jar file, then each should have it listed as a single mount point. When both factories are closed, then the mount point can be unmounted. How many resources that a given factory creates against its single mount is irrelevant to the reference counting for the close.

Fundamentally, the number of mount points in the factory for:

   Resource foo = factory.newResource("jar:file:///tmp/test.jar!/foo/");
   Resource bar = factory.newResource("jar:file:///tmp/test.jar!/foo/bar.txt");

should be identical to

   Resource foo = factory.newResource("jar:file:///tmp/test.jar!/foo/");
   Resource bar = foo.resolve("bar.txt");

The count of mounts must be independent of how many resources are created against it.

The problem is in mountIfNeeded(URI uri) as it is always mounting, even if not needed. It should be checking existing mounts and not mount again if it is already mounted.

[gregw]

@joakime I have pushed a quick fix:

• We were already only mounting the root of a jar, but now we also record that as the root Resource
• mountIfNeeded now checks existing mounts to see if we have already mounted the root Resource. There is probably a much better way of doing this than by string comparison.
• if we have already mounted the root of a jar URI, then we return null from mountIfNeeded as another mount is not needed.
• we now don't need special handling for non existent resources. The root is still mounted and the non existent path is forgotten

This almost certainly needs more work/optimisation/cleanup, but I hope it conveys what I think needs to be done.