Adventure Mode v0 (E0-E6): ramificare per-raspuns in campanie

Flag opt-in `adventure` (default off) — zero regresie non-adventure.
E0: `adventure:false` in defaultState + checkbox builder + `ADVENTURE` in orchestrator.
E1: `_lastGiven` in libJS; `checkAnswer` captureaza raspunsul; `campaignDone`
    calculeaza cheia branch ('*'/text-tf/index-choice) si o adauga in payload nextRoom.
E2: `resolveBranch(idx,key)` + rutare nextRoom: 'end'→owExitUnlocked+showOverworld;
    numar→owUnlocked[dest]+showOverworld(dest). Non-adventure: comportament existent.
E3: `owCheckEnter` blocheaza usi incuiate (ADVENTURE&&!owUnlocked); exit via
    owExitUnlocked. `owRefreshDoors`: stil `.locked` (dim+lock). `__ow.state`:
    expune owUnlocked/owExitUnlocked.
E4: `saveProgress` adauga doneList+owUnlocked+owExitUnlocked+target; `tryResume`
    reconstruieste din doneList non-contiguu (nu bucla liniara 0..idx).
E5: `buildDiploma`: ADVENTURE&&!roomDone[i] → 'neexplorata' (nu stele inselatoare).
E6: Builder UI — `normalizePuzzle` garanteaza p.branch={}; `cleanState` clampa
    tintele+strip branch cand !adventure; `puzzleCard` afiseaza dropdown-uri
    ramificare per-puzzle (free=1, tf=2, choice=1/optiune); `data-fb` handler;
    `adventure` change → renderPuzzles().

Smoke 35/35 (4 teste noi: branch-jump, resume non-contiguu, regression, tf branch).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

AGENTS.md

@@ -20,9 +20,9 @@ sursa de adevăr tehnică pentru agenți.
 python3 -m http.server 8000
 
 # Teste (Playwright; fără package.json commitat — vezi tests/AGENTS.md):
-npx playwright test tests/smoke.mjs                    # suita completă: 31/31
+npx playwright test tests/smoke.mjs                    # suita completă: 35/35
 npx playwright test tests/smoke.mjs --grep @regresie   # regresie: 16
-npx playwright test tests/smoke.mjs --grep @campanie   # campanie E2E: 17
+npx playwright test tests/smoke.mjs --grep @campanie   # campanie E2E: 21
 ```
 
 ## Durable Rules (repo-wide)