Fix memleak for long-lived resultsets.

[vovkasm]

It is a recursive attributes hash chain when call search on the same
resultset many times. It looks like pseudo-leak with long-lived resultsets.

In our application (that long-lived daemon) we use something like this:

sub periodic_refresh {
  state $rs = $schema->resultset('ServerBackend')->search({status=>'active'},{order_by=>'me.id',prefetch=>'Server'});
  my @fresh_backends = $rs->all; # now we use this workaround: $rs->search; but it is slower because new resultset created on every iteration
  # other refresh logic
}

This patch fix a problem.
I'm just not sure of a good name for the test file. Now it is 110leaktrace.t, but it can be renamed to something more appropriate.

[ribasushi]

Saw it, bug is real. Will be looked at later today (I believe there are more issues of this nature)

[ribasushi]

This finally landed in master as 6ae62c5c1, though with a both different test and different solution. Apologies for the delay, the tooling needed to be improved before asserting this is indeed the end of the issue.

Cheers!