feat(devtools): rank dag_export --post-asap without a cost document

[Selvomega]

--post-asap without --planner-cost-json ran nothing. The gate exists because a post-ASAP export asserts a cost-ranked decision, and dag_export owns no deployment evidence with which to defend one. But a user with no deployment yet still wants to see which replacements the search finds, which one it commits to, and what the merged post-ASAP DAG looks like — and that needs the structure to be real, not the numbers.

Add --default-cost, mutually exclusive with --planner-cost-json. It ranks candidates with DefaultCostModel and keeps export_model as None, so every annotation stays CostSource::Unavailable with no value, workload_cost_summary degrades to an unavailable summary, and the viewer renders "Not estimated". The structural number decides which candidate wins and is then discarded, never serialized. The flag pair is rejected rather than given a precedence order, before the document is parsed, because the two disagree about what the export may claim, not just about which model ranks. --topk-margin-json still applies under --default-cost: accuracy evidence is orthogonal to cost evidence.

Absent both flags the export is byte-identical to before; only the stderr line differs, which now names both cost sources instead of demanding evidence.

This is not a revival of what 5dbf4d9 removed. That commit deleted two things: winner_cost_annotations emitting a modeled CostAnnotation from AnalyticalCostModel, and the analytical.map_or(&default_model, ...) choice of which model ranks. Only the latter returns. A DefaultCostModel number was never an exported cost, and structural node counts remain intentionally unusable as a cost fallback.

Closes #431