loading of the class list is too slow about javadoc-reader HOT 2 CLOSED

Loading is much faster now (it is typically <1s) so I'll close this issue.

The relevant changes are:
1. The "index" computed on the server sorts the packages and the classes so the
client doesn't have to.
2. The index is smaller. The names of the packages are not repeated, but rather
pointed to in the class list.
3. The index is cached (for 1h now) on the server so that often the javadoc 
page that
lists the classes on the foreign server is not loaded.
4. The client uses arrays where it used to have TreeSet and later HashSet. The
TreeSet seems terribly slow in javascript. And anyway, the uniqueness was making
javadoc-reader incorrectly handle classes with the same name in different 
packages.

Finally, this is more a usability issue: Since the number of results displayed 
by
default to 20 for performance reasons (plus, I expect people not to want to 
wade thru
huge lists anyway) there was a need for requesting more results. Now there is a
button to do so.