admin: decide what the overview's failure count means #647

After #641, the overview's top_failure_categories[].affected_users means three different things depending on the range and the filters (query_service.py:951-978):

• Past days, no dimension filter: the sum of each day's distinct-user affected_users rollup. A user who fails on 5 days counts 5 times, so this is user-days.
• Today: distinct users in today's raw slice, added on top.
• Billing/provider/model filter set: event_count rows, so it counts events. The PR body notes this case.

On main it is distinct users over the whole range, from one raw read (query_service.py:750-760). The page shows it in a column labelled "Count".

Options:

1. Keep the computation and rename the field (and label) to what it counts.
2. Return distinct users over the range. The per-day rollups can't give that: it needs a raw-event read (bounded by the 180-day retention) or a new rollup.
3. Count events everywhere: sum event_count and drop the affected_users path.

The activation funnel has the same shape (query_service.py:933): it sums the per-day distinct activation_users rollup across days. Whatever is decided here should apply there too.

Owner decision. Depends on #641.