from slicename(), which breaks on names with a dash in them.

that loads one of five different modules, based on the PATH_INFO.  The
modules do nothing but pull in the proper library; all of the xmlrpc
and authorization code is in one place now (the wrapper), instead of
being duplicated. Much cleaner, much safer.

An update to httpd.conf is required, and done with update script 7.

Takes old certificate, gets the private key from inside it,
generates a new serial number, and uses x509 to create a new
certificate. Good for updating expired certs.

nodes is selected, those links must use vlan encap. Since shared node
selection is not done until late, we end up goaing back and modifying
the encap type and forcing a vlan to be created.

This is done cause the shared nodes (openvz) do not support veth
encap, and so arp causes leakage between experiments on the same node
that have common IP addresses.

are two ways this can happen.

1. When allocating the proxy nodes before getting the tickets. This is
   just like the local node case; need to go around the loop again and
   retry with a new node set. 

2. When getting the tickets, someone else got the nodes and gthe
   getticket fails. This is a lot like case 1, but now we have to
   nfree the proxy node, and tell the mapper to try again.

attempts by protogeni code.

The toolchain build should be dependent on an actual file to ensure we
don't rebuilt it for every target that goes into the initramfs.

Also rename old fake tcsh to tcsh.fake, and run a new script that
symlinks tcsh and csh to tcsh.fake if tcsh doesn't exist.  This way tcsh
can be removed from the build but tcsh/csh users can still log in.