Bogus cleanup code in PostgresNode.pm

I noticed that even when they are successful, buildfarm members bowerbird
and jacana tend to spew a lot of messages like this in their bin-check
steps:

Can't remove directory /home/pgrunner/bf/root/HEAD/pgsql.build/src/bin/scripts/tmp_check/data_main_DdUf/pgdata/global: Directory not empty at /usr/lib/perl5/5.8/File/Temp.pm line 898
Can't remove directory /home/pgrunner/bf/root/HEAD/pgsql.build/src/bin/scripts/tmp_check/data_main_DdUf/pgdata/pg_xlog: Directory not empty at /usr/lib/perl5/5.8/File/Temp.pm line 898
Can't remove directory /home/pgrunner/bf/root/HEAD/pgsql.build/src/bin/scripts/tmp_check/data_main_DdUf/pgdata: Permission denied at /usr/lib/perl5/5.8/File/Temp.pm line 898
Can't remove directory /home/pgrunner/bf/root/HEAD/pgsql.build/src/bin/scripts/tmp_check/data_main_DdUf: Directory not empty at /usr/lib/perl5/5.8/File/Temp.pm line 898
### Signalling QUIT to 9156 for node "main"
# Running: pg_ctl kill QUIT 9156

What is happening here is that the test script is not bothering to do an
explicit $node->stop operation, and if it doesn't, the automatic cleanup
steps happen in the wrong order: the File::Temp destructor for the temp
data directory runs before PostgresNode.pm's DESTROY function, which is
what's issuing the "pg_ctl kill" command. On Unix that's just messy,
but on Windows it fails because you can't delete a process's working
directory. I am not sure whether this is guaranteed wrong or just
sometimes wrong; the Perl docs I can find say that destructors are run in
unspecified order once interpreter shutdown begins. But by adding some
debug printout I was able to verify on my own machine that the data
directory was already gone when DESTROY runs.

I believe we can fix this by forcing postmaster shutdown in an END
routine instead of a DESTROY routine, and hence propose the attached
patch, which does things in the right order for me. I'm a pretty
poor Perl programmer, so I'd appreciate somebody vetting this.

regards, tom lane