[pattgen,dv] The cnt_rollover test doesn't really work

### Description

As I understand it, this was supposed to be setting large `prediv` values and checking the counter worked properly. A good idea! Unfortunately, it doesn't actually sample the enormous 32-bit range for the register. (Maybe the spec changed over time)

Waiting 4 billion cycles is not a realistic approach(!) so we will need to force a value to be "near the top" during the simulation.

Reading that carefully, I also realised that the coverage tracking for this in `pattgen_cov_if` doesn't really make much sense. I think the author was explicitly looking to see the rollover, but does so by looking at the threshold values instead of the counters. Oops!

Locate the cnt_rollover test and the pattgen_cov_if coverage logic in the pattgen/dv area, then read how the counter and threshold values are sampled. Done means the simulation exercises a counter value near the 32-bit maximum without waiting billions of cycles, and coverage tracks the actual rollover behavior.

Written by the indexing model from the issue text.