fix: research unblock adapters after #75/#76 (bench def + progress merge)

[jkbennitt]

Why

#75 seeded a built SimpleResearchBench at (128, 0, 136) and queued currentProj=Smithing. #76 pin-matches native AppData before game/load. Re-smoke after #76 still had research_bench_present: False and research_target Smithing locked (misleading “bench/prereqs” text). The save was correct; RLE adapters were not.

RCA (verified against RIMAPI DTOs)

1. BuildingDto uses snake_case def. _building_def_name only read def_name then label → "simple research bench". is_research_bench_def stripped _ only, so spaces broke the researchbench substring match. The seeded bench was dropped from the 50-structure sample / brief flag.

2. get_research() only read /research/summary. That DTO has counts + by_tech_level bags, not current_project. The adapter treated any tech level with finished > 0 as entirely completed, then truncated to finished_projects_count. Medieval (incl. Smithing) landed in completed, got truncated out of available, and the fix: unblock Crashlanded research scoring (seed bench + brief observability) #75 validator reported locked. Seeded Smithing never hit the already-satisfied path.

Live shapes (IlyaChichkov/RIMAPI): BuildingDto.Def / Label; /research/progress is ResearchProjectDto (name, progress_percent); /research/finished is finished_projects; /research/tree has is_finished / can_start_now.

What landed

• Map building field def → def_name → label.
• Strip spaces (and keep underscore strip) in is_research_bench_def.
• Merge /research/progress (current + percent) with /research/finished + /research/tree so tick0 current can be Smithing and available/completed are honest (can_start_now, not by-tech-level bags).
• Drop the broken summary bucketing. Mock payloads that already have current_project still pass through; missing extra endpoints are optional.

Tests (unit only)

• Building def field → bench matcher true; get_map keeps SimpleResearchBench when live JSON uses def + spacey label.
• Spacey label ("simple research bench") matches.
• Progress merge exposes current Smithing; available/locked/finished status is honest (Smithing is current, not locked).

Out of scope

• No scenario-save re-seed
• No scoring formula / SCORING_VERSION bump
• No matrix, smoke, or billed model/cal runs

  

[jkbennitt]

LGTM — RCA matches the post-#75/#76 smoke miss: BuildingDto def (+ spacey label strip) so SimpleResearchBench stays in the sample/brief; research adapter merges /progress+/finished+/tree instead of broken by-tech-level summary bucketing so Smithing is current not locked. Unit tests cover live def, spacey label, progress merge + research_target_status. No scoring/save re-seed/matrix. All five CI SUCCESS on 84660a4 (run 34264192479). HOLD merge for CoS clear; plain squash (no --admin).