The server was not cleaning the last IO error and error number when
resetting slave.

This patch addresses this issue by backporting into 5.1 part of the
patch in BUG 34654. A fix for this issue had already been pushed into
6.0 as part of the aforementioned bug, however the patch also included
some refactoring. The fix for 5.1 does not take into account the
refactoring part.

mysql-test/extra/rpl_tests/rpl_reset_slave.test:
  Backported the test case and improved with deploying include/start_slave.inc
  in relevant spots.
sql/slave.cc:
  Backported part of patch from 6.0 that includes cleaning 
  mi->clear_error() at:
    1. beginning of handle_slave_io
    2. on successful connection
  
  Also, backported the assertion added in the original patch.
sql/sql_repl.cc:
  Backported the call to mi->clear_error() on reset_slave().