feat(media): extend Telegram bot support to documents, audio, voice, and video

[Martins6]

Summary

Previously DefaultHandler in internal/bot/handlers.go only processed photo attachments; documents, audio files, voice messages, and videos were silently dropped even though their media types were already detected. This PR routes all five attachment types through a single media-processing path that mirrors the existing photo schema (download from Telegram → save under <workspace>/downloads/<type>/<UTC-timestamped-filename> → forward a prompt that references the local file to OpenCode).

Changes

• internal/media/downloader.go

  • Add MediaMetadata struct (Kind, FileSize, MimeType, OriginalName).

  • Rewrite BuildPrompt(path, meta, caption) to emit a metadata-rich prompt:

    File located at: <abs-path>
    File type: <photo|document|audio|voice|video>
    File size: <bytes> bytes
    Original name: <name>      ← only when Telegram provides FileName
    MIME type: <mime>          ← only when Telegram provides MimeType
    
    User message: <caption>
    

  • Add ExtractFileRef(message) returning (fileID, fileSize, mimeType, originalName, mediaType, ext, ok) for any recognized attachment.

• internal/bot/handlers.go

  • Extract the photo-only branch into a private processMediaAttachment helper that handles all five media types through one path.
  • Add mediaKindLabel for capitalized log labels per attachment kind.
  • Simplify the empty-message skip guard to rely on the resolved userMessage.

• internal/media/downloader_test.go

  • Update 2 existing BuildPrompt tests for the new signature.
  • Add 11 new tests covering metadata-rich prompts (full / partial / zero-size) and ExtractFileRef for each media type plus the empty-message case.

• docs/features/feat-telegram-bot.md — describe the new behavior and prompt format.

• docs/features/LOG-FEATS.md — log entry.

• docs/features/OVERVIEW.md — refresh the file description.

Behavior

• Photos, documents, audio files, voice messages, and videos are all downloaded from Telegram and saved under <workspace>/downloads/<type>/<UTC-timestamped-filename>.
• The prompt to OpenCode now includes file type, size, and (when Telegram provides them) the original filename and MIME type.
• Photos gain a metadata block in their prompt; everything else about photo behavior is preserved.
• Stickers, contacts, and locations remain unsupported.
• No explicit size limits added; relies on Telegram Bot API's 20 MB cap and the existing 30 s HTTP timeout.

Verification

• go build ./... clean
• go vet ./... clean
• go test ./... all packages green (20 tests in internal/media/...)

Test plan (requires a running bot + OpenCode CLI)

1. Send a photo with caption → saved under downloads/images/, prompt contains File type: photo, File size: <N> bytes.
2. Send a PDF document → saved under downloads/documents/, prompt contains Original name: …pdf, MIME type: application/pdf.
3. Send a voice message → saved under downloads/audio/ with .ogg.
4. Send an MP3 audio file → saved under downloads/audio/ with .mp3.
5. Send a short video → saved under downloads/videos/ with .mp4.
6. Regression — plain text message still works; empty/whitespace-only message still skipped.