Refine Scenario configuration and sizing

[Roman Lutz (romanlutz)]

Summary

Adds named technique sets, Adaptive attempt caps, searchable datasets, per-dataset limits, compatibility-aware identities, and exact/ranged backend-owned run arithmetic. Default estimates use bounded concurrency with execution-only timeouts.

What this layer adds

• Named technique sets plus explicit Custom selection.
• Adaptive per-objective attempt limits and baseline comparison controls.
• Searchable dataset selection with bounded per-dataset caps and provenance.
• Compatibility-aware identities and unsupported Jailbreak-combination filtering.
• Backend-owned exact/ranged run-size arithmetic shared by preview and launch.
• Bounded concurrent default estimation with queue time excluded from execution timeout.

Preview

Stack

This is 6 of 7 in native GitHub Stack #2390 and depends on #2376.

Layer	Scope	Pull request
1	Progress foundation	#2372
2	Catalog and launch	#2373
3	Live progress	#2374
4	Run history	#2375
5	FIFO scheduling	#2376
6	Configuration sizing (this PR)	#2377
7	Result details	#2378

See the full implementation, screenshots, walkthrough, validation, and operational notes.

Validation

• Backend: 15,212 passed, 120 skipped.
• Frontend: 68/68 suites, 1,286/1,286 tests; lint, TypeScript, and production build passed.
• Mock E2E, retries disabled: 91 passed, 0 failed, 0 skipped.
• Seeded E2E, retries disabled: 62 passed, 0 failed, 0 skipped.
• All configured pre-commit hooks passed; distinguished-engineer review approved the final stack.

Compatibility notes

Unsupported Jailbreak v4 technique combinations are rejected, v3 persisted Jailbreak runs intentionally do not resume as v4, and configuration-aware Adaptive identities may regroup historical analytics.