A bug in the CPUSTAT table entry generation code meant that the _1sec and _20sec
tables did not maintain history correctly.

This is fixed, and a new ndbinfo_cpustat test is added to cover this and
potentially other cpustat testing yet to come.