planner: deploy DDSketch from MILP plans #802

Follow-up to A3 of #790. The engine has a DDSketch aggregation (#792), but the MILP never plans one: sketch-bench's `dd` variant isn't in `DEPLOYABLE_FAMILIES`. To deploy it:

• sketch-bench: add `dd` to `DEPLOYABLE_FAMILIES`.
• ASAPQuery (`optimizer/milp_output.rs`): map `("dd", Quantile)` → `AggregationType::DDSketch` with `alpha` = `sketch_config.params.alpha` (same key on both sides). Land this together with the sketch-bench change.
• Memory sizing: ASAPQuery sets `MetricFacts.value_range = None`, so DDSketch is costed at its measured memory instead of being sized to the metric's real value range. Add a per-metric value range to the workload facts file.
• Accuracy metric: Quantile Raqes require `max_rank_err ≤ 1 − sla`. DDSketch guarantees relative value error (reported since Report DDSketch relative value error sketch-bench#164), so a rank-error check judges it on a metric it doesn't optimize for. Decide how a quantile query states which error it needs.

🤖 Generated with Claude Code