v0.5.1 — Telegram Markdown formatting

Patch release. Three packages bumped in lockstep and published to npm with provenance:

• bazilion@0.5.1 — CLI + bundled daemon
• @bazilion/client@0.5.1 — hermetic HTTP client
• @bazilion/api-types@0.5.1 — zero-dep wire types

What's new

Telegram output now reaches parity with the Web UI's Markdown rendering. Strictly additive — only the agent's own replies are affected.

• Markdown formatting on Telegram. Agent replies used to arrive as plain text, so **bold**, # headings, and `code` rendered as literal syntax. They're now converted to Telegram HTML using the same marked parse the Web UI uses — bold, italic, strikethrough, inline code, code blocks, links, and blockquotes map 1:1.
• Sensible approximations for what Telegram can't render. Telegram's formatting is a fixed entity set, not arbitrary HTML, so: headings → emoji-prefixed bold (▎/▸/· by level, since Telegram has no font sizing); tables → an aligned monospace block; lists → •/1. bullets with indentation and ☐/☑ task items; --- → a horizontal rule.
• Long replies are split, not truncated. A tag-aware splitter chunks replies over Telegram's 4096-char limit into multiple messages, closing and reopening any entity that straddles a boundary so every chunk is valid on its own.
• Never drops a reply. If Telegram rejects a chunk's entities, the mirror retries that chunk once as plain text. Error and verbose tool-trace lines continue to send as plain text.