refactor: migrate main.py to modular routers and add project roadmap

- Migrated monolithic main.py to feature-scoped routers in app/routers/
- Added GEMINI.md for project context and AI instructional guidelines
- Updated README.md with a comprehensive modernization plan (SQL migration, robust scraping DSL, frontend modernization)
- Improved authentication with cookie support and modular JS
- Updated test suite and documentation

app/downloaders/video_players/AGENTS.md

@@ -0,0 +1,37 @@
+# Video Players (app/downloaders/video_players)
+
+## OVERVIEW
+File hosting extractors that extract direct download links from video player pages (Doodstream, Sibnet, VidMoly, etc.).
+
+## WHERE TO LOOK
+
+| Need | File |
+|------|------|
+| Base class | `base.py` - `BaseVideoPlayer` abstract class |
+| Add new player | Create new `.py` file, inherit `BaseVideoPlayer`, add to `__init__.py` |
+| URL detection logic | Each player's `can_handle()` method |
+| Extract download link | Each player's `get_download_link()` method |
+
+## CONVENTIONS
+
+**Class naming**: `{Provider}Downloader` (e.g., `DoodStreamDownloader`)
+
+**Required methods**:
+```python
+def can_handle(self, url: str) -> bool: ...
+async def get_download_link(self, url: str, target_filename: str = None) -> tuple[str, str]: ...
+```
+
+**File operation**: Always use `sanitize_filename()` on extracted filenames.
+
+**HTTP client**: Use `self.client` (AsyncClient from base class). Always close via `await self.close()` when done.
+
+**Return format**: `(download_url, filename)` tuple.
+
+## ANTI-PATTERNS
+
+- Do NOT hardcode User-Agent in each player (use base class headers)
+- Do NOT forget to call `await self.close()` after extraction
+- Do NOT return None for missing URLs, raise an exception
+- Do NOT use sync `requests`, use async `httpx`
+- Do NOT skip the `target_filename` parameter, even if unused