Suppression of small sample sizes in filters and results (minimum group-size threshold / deidentification)

Nobody has claimed this yet.

**What problem will this feature request solve?**
In people/HR analytics — and equally in social science, behavioral science, public-health, education, census, and other sensitive population research and analysis — exposing data for very small groups can re-identify individuals. This is the classic **small-cell / small-sample suppression** requirement from statistical disclosure control and **deidentification** (a.k.a. de-identification, k-anonymity / minimum-cell-size rules): values backed by fewer than *N* records must be **suppressed** so individuals can't be inferred.

Today there is no built-in way to enforce a minimum-group-size suppression threshold in Metabase. Two gaps in particular:

- **Filters:** linked filters cascade correctly (parent → child), but there is no way to require a minimum underlying row count for a value to appear in a dropdown. Small groups (e.g. fewer than 5 users) still show up as selectable options.
- **Results:** the same small groups still appear in chart results — e.g. every demographic value becomes its own line in a time-series, including a group of 3 users.

Workaround attempted: pointing the child filter at a saved question with a `HAVING COUNT(...) > 4` clause ("From another model or question" as dropdown source) *does* apply the count threshold, but doing so **bypasses linked filters** — the child dropdown no longer respects the parent selection (e.g. the demographic-value list stops narrowing to the chosen demographic-type). So today a user can get *either* suppression *or* cascading linked filters, but not both.

**Describe the solution you'd like.**
A first-class way to apply a minimum-count **suppression threshold** to a linked filter and the charts it drives, so values whose underlying row count is below the threshold are suppressed from:
- the filter dropdown options, AND
- the chart results (e.g. a per-group line is omitted entirely from a time-series when that group is below the threshold).

The threshold is a count of underlying rows/users (not a count of distinct options). One suggested shape: allow a saved question with a `HAVING`-style aggregate condition to serve as a linked-filter dropdown source **while still respecting parent linked-filter selections**. For below-threshold groups in results, the preferred display is to omit them entirely (or show a "Below privacy threshold" placeholder) rather than showing `0` or the real count — showing `0` falsely implies "no activity."

**How does this feature request impact you?**
High — without this, teams cannot safely build the analytics dashboards they need; suppressing sub-threshold groups is a hard privacy/compliance requirement across HR/people analytics, social and behavioral science, public health, and other sensitive population research, not a preference.

**Additional information**
n/a

No files or tests are named. Start by tracing how linked-filter options are queried and how grouped chart results are produced, then identify where parent selections and aggregate thresholds could be preserved together. Done means below-threshold groups are absent or clearly suppressed in both dropdowns and chart results without breaking cascading filters.

Written by the indexing model from the issue text.