[dv, fcov] Look at use of U mode in riscv_mem_intg_error_test

[GregAC]

We're rarely hitting some bins of priv_mode_irq_cross, though associated with internal NMIs seen in U mode (which occur on ECC errors), increasing test iterations does help but doesn't reliably hit them every regression (#2051)

Investigate why and rework the test to more reliably cover everything without excessive iterations.