Easy to prevent crash, but the question was how to treat this case?
  We ended up implementing the "global" SPs (i.e. with no associated
  db), which were planned but left unresolved when SPs moved into dbs.
  So now things like "call .p()" work too.


mysql-test/r/sp-error.result:
  New test case for BUG#5000, and "global" SPs in general.
mysql-test/t/sp-error.test:
  New test case for BUG#5000, and "global" SPs in general.
sql/sp.cc:
  Prevent crash when the new db is null.
sql/sp_head.cc:
  Don't set the db part of the name to thd->db, we have already set it correctly
  in the provided name struct.
  Also, don't attempt to change "no-db" when executing an SP.
sql/sql_yacc.yy:
  Added support for the "global SP" syntax, e.g ".p()".