Regression test pipe8 causes mismatches if run on Thor with multiple slaves per node.

The test involves a PIPE write and PIPE reads, that will run on all slaves.
The write, writes to a fabricated temporary physical file name, which is constant across the slaves.
So the write on each will be trying to write to exactly the same physical file.
And the PIPE read will all read the same file.
On 3 a system with slaves on different physical nodes you don't notice (or on a 1-way), but on a n-way on the same node, the writes will overlap and the reads will all read the same physical file.

This results in a mismatch in the test, where either there are too many rows read (read multiple times by each slave), or potentially no rows read, because one of the slaves that had no rows to write, overwrote the slave that did.

Fix by ensuring test writes to a unique (per slave) physical file name.