feat(telegram-remote): ship lib/ code, add voice transcription, chat pagination, stream photos

- Add lib/ (plain-ESM plugin code, no build step) so the published
  tarball actually contains the plugin; un-ignore lib/ for this package
- Sync README: voice/audio transcription via home-lab Whishper (default
  large-v2, device cuda, language auto), /chats pagination and subagent
  hiding, whisper* config keys, updated security note

packages/telegram-remote/README.md

@@ -9,6 +9,7 @@ Turn any Telegram chat into a pocket terminal for your harness: chat with your A
 ## ✨ Features
 
 - 💬 **Chat from Telegram** — send a message, get a live-streamed structured reply (thinking, tools, final answer) edited in place
+- 🎙️ **Voice & audio** — send a voice message or audio file; it is transcribed through your home-lab Whisper server (default `large-v2` on CUDA GPU) and the text is sent to your AI
 - 🎛 **Full harness control** — 49 commands: chats, models, sessions, subagents, goals, jobs, files, PowerShell, exports, presets, skills, settings, credentials, permissions, and more
 - 📡 **Live state** — status, running turns, queued messages (steer / edit / remove), background jobs
 - 🔐 **Permission-aware** — sandbox read/write/full control per chat, approval buttons for risky tools
@@ -77,6 +78,12 @@ Turn any Telegram chat into a pocket terminal for your harness: chat with your A
 | `allowEval` | `true` | Enable `/eval` (runs JS in the harness) |
 | `notifyOnStartup` | `true` | Send a "bot online" message on boot |
 | `stateFile` / `logFile` | `~/.dsh/...` | Override runtime state / log paths |
+| `whisperBaseUrl` | `"http://192.168.31.159:8082"` | Home-lab Whishper server base URL |
+| `whisperModel` | `"large-v2"` | Whisper model used for transcription |
+| `whisperDevice` | `"cuda"` | Whishper device (`cuda` or `cpu`; GPU is `cuda`) |
+| `whisperLanguage` | `"auto"` | Language hint (`"auto"` = auto-detect; e.g. `ru`, `en`) |
+| `whisperTimeoutMs` | `120000` | Max wait for a transcription to finish |
+| `whisperMaxBytes` | `20971520` | Max accepted audio size in bytes (20 MiB) |
 
 ## 💬 Usage
 
@@ -116,7 +123,7 @@ Tap the **/start** keyboard or type `/help` for the full command list. Key comma
 - The bot can control your computer — **only add users you trust** (`ownerChatId` / `allowedUserIds`).
 - Every request is checked against the per-chat `/permission` mode; risky tools request approval.
 - `/eval` and `/cmd` are powerful — consider `allowEval: false` unless you need them.
-- No telemetry, no external services: the bot talks to Telegram and your harness only.
+- No telemetry: the bot talks to Telegram, your harness, and (for voice/audio) your home-lab Whishper server only.
 
 ## 🔄 Hot reload (development)