feat: preserve message entities as markdown

[ohld]

Summary

• Added message_to_markdown() function that converts Telethon message entities to markdown format
• Fixed UTF-16 offset handling for emoji characters (Telegram uses UTF-16 code units, Python uses Unicode code points)
• Updated all 10 message retrieval functions to use the new converter

What changed

Previously, all message retrieval functions (get_messages, list_messages, search_messages, get_history, get_pinned_messages, get_chat, get_last_interaction, get_message_context, format_message) used msg.message which returns plain text only. All Telegram formatting entities were lost:

• Hyperlinks (MessageEntityTextUrl) — visible text returned, URL lost entirely
• Bold (MessageEntityBold) — formatting stripped
• Italic (MessageEntityItalic) — formatting stripped
• Strikethrough, code, pre — all stripped

Now entities are converted to markdown:

• MessageEntityBold → **text**
• MessageEntityItalic → _text_
• MessageEntityStrike → ~~text~~
• MessageEntityCode → `text`
• MessageEntityPre → ```lang\ntext\n```
• MessageEntityTextUrl → [text](url)
• MessageEntityMentionName → [text](tg://user?id=X)
• MessageEntityUrl, MessageEntityMention, MessageEntityHashtag — kept as-is (already visible in text)

UTF-16 fix

Telegram entities use UTF-16 code units for offsets. Characters outside the BMP (most emoji like 👨, 🔗, etc.) take 2 UTF-16 code units but only 1 Python character. Without conversion, entity boundaries shift and formatting appears mid-word (e.g., с**вязка** instead of **связка**).

Added _utf16_to_python_offsets() to correctly convert between the two coordinate systems.

Test plan

• Unit tests with plain text (no entities)
• Unit tests with bold, italic, links
• Unit tests with emoji characters (UTF-16 surrogate pairs)
• Tested on real Telegram channel with ~30 posts
• Verified hyperlinks preserved in real posts

🤖 Generated with Claude Code