feat(notes): add /api/notes/gdoc and /gdoc/edit proxy routes

feat(notes): add /api/notes/wolfram proxy route

fix: ?dateien=1 filter time (single NER pass, 420 s timeout, clearer errors)

style: release page with cards, dark mode, select-all

feat: release page lists asked files, multi-select, code remembered on the device

feat: /iserv/freigabe releases unchecked files (APPROVE_TOKEN, 10 min, once)

feat: /iserv/tasks?dateien=1 passes original files without names

fix(iserv): no repeated text from merged DOCX cells

feat(iserv): shorter output, finished tasks only with ?alle=1

feat(iserv): filtered IServ tasks for Muse

feat(notes): stream chat/completions via SSE passthrough

fix(notes): pin Gemini requests to Google AI Studio provider

fix: only disable reasoning for non-Gemini models in Notes proxy

fix: pass client-chosen model/fallback chain through Notes proxy

fix: read Tavily key from TAVILY_API_KEY_2

fix: require tool-call support from routed provider, add Tavily search proxy

Savestate: Fix Hugging Face health-check binding

fix: restore Authorization Bearer template literal in Notes proxy

fix: restore template literal in OpenRouter proxy error response

Savestate: Fix Dockerfile and add Browser_Agent for HF build, bump version to 1.8.334

Savestate: Increment version to 1.8.333 for HF deployment

Savestate: Merge Hugging Face main changes (Notes AI proxy, 127.0.0.1 binding) into master

Savestate: Update server.js before deployment

Savestate: Update DeepSeek v4 Flash ID to static 0731 version to resolve OpenRouter 400 error

Savestate: Update model display names to DeepSeek v4 Flash and Pro

Savestate: Set DeepInfra as the exclusive provider for DeepSeek v4 Pro due to privacy and price-cache optimum

Savestate: Revert DeepSeek v4 models to use -latest suffix and re-enable ZDR data_collection deny for cache discount security

Savestate: Update DeepSeek v4 config for caching and disable ZDR

Savestate: Add ZDR and OpenRouter specific routing for DeepSeek models

Savestate: Add OpenRouter DeepSeek V4 models and increment build version

fix: use env in sudo to properly pass variables

fix: pass PLAYWRIGHT_BROWSERS_PATH to agent_runner via sudo

fix: pass OPENROUTER_API_KEY in start.sh

feat: add SCHOOLMIND prefix to logs

Feat: Implement invisible pending queue for quick settings modal to allow background processing without blocking UI

Fix: Stop dispatching flashcards-updated on every single iteration of retroactive enrichment

Feat: Run AI correction and enrichment directly inside the vocabulary modal flow before saving to the database

Revert fallback deck logic: strictly use Schnelleingabe deck as requested

Fix: Correct db update method, iterate all decks, respect user's custom deck

Fix: Revert back to gemini-3.1-flash-lite as requested by user

Fix: Retroactively enrich old vocabulary cards lacking explanations, fix Gemini model name

Feat: Auto-show keyboard when vocabulary modal opens via Capacitor Keyboard Plugin

Feat: Add background typo correction via AI in enrichment queue

Feat: Add DeepL Free API integration for instant translation with background AI enrichment using Gemini 3.1 Flash Lite

Perf: Shorten overlay success screen timeout to 1s

Perf: Fast overlay startup - skip auth/sync init, hide body until ready

Fix: Call getCardsForDeck instead of getFlashcards

Feat: Prevent adding duplicate vocabulary words in Schnelleingabe

Feat: Auto-correct user input in Schnelleingabe if AI detects a typo

Fix: Eliminate black flash on overlay open by hiding window until JS signals overlayReady