Stream tokens to the terminal during `write` (Rich live output)

[telivity-otaip]

Why

write currently waits for the full generation before printing. Streaming makes the CLI feel alive, helps users catch slop early, and is a classic high-visibility contribution for a terminal product.

Scope

• Stream MLX generation tokens to stdout while drafting (--channel post first; article sections can stream per section)
• Keep --json non-streaming (emit one final receipt) unless --json --stream is explicitly designed
• Show a small Rich status line: channel, adapter=none|path, section N/M for articles
• Ctrl-C should abort cleanly without corrupting profile state

Pointers

• src/personality_protect/write.py — generation
• src/personality_protect/write_article.py — section loop
• src/personality_protect/mlx_runtime.py — MLX calls
• cli.py write command

Acceptance criteria

• Post channel prints tokens as they arrive on a TTY
• Article channel streams each section (clear section separator, no markdown chrome)
• --json remains a single final object (backward compatible)
• Piped/non-TTY output either buffers safely or documents behavior
• Tests cover the non-MLX / mock path (callback invoked with chunks)
• No personal corpus in fixtures — Contoso only

Difficulty

Easy–Medium — mostly plumbing a stream callback through existing generate helpers.