feat(tui): count blocks arriving below a scrolled view in the follow hint

[Max17190]

Why

A reader scrolled up has no way to know whether anything landed at the bottom; the hint says only how to get back. Forced or silent follow behavior is the largest complaint class in comparable terminal frontends.

Summary

Count blocks appended while the view is scrolled up and render the count ahead of the re-arm hint (N new · esc follow · pgup/pgdn scroll). The count clears whenever the view reaches the bottom, so a later scroll-up starts from zero instead of resurrecting content already seen. Chat dirt promotes a chrome repaint while a count shows, so the status line never goes stale.

Test Plan

• unread_counts_blocks_while_scrolled_and_clears_at_the_bottom (transcript state machine, incl. no-resurrect).
• scrolled_hint_counts_blocks_arriving_below (hint text, chrome promotion, no-resurrect at the app layer).
• Both shown red with the counter neutered. Workspace tests and clippy clean.

Greptile Summary

Adds an unread-block count to the follow hint when completed transcript blocks arrive below a scrolled view. Returning to the latest content clears the count, and status chrome continues to repaint through the existing chat invalidation path.

Confidence Score: 5/5

No blocking failure remains.

Focused checks confirmed unread blocks increment while the view is scrolled, clear when the view reaches the bottom, and appear in the status hint. The full TUI test suite also completed successfully with 248 passing tests and 6 ignored tests.

T-Rex Logs

What T-Rex did

• Ran focused Rust tests covering unread block counting, unread clearing on return to bottom, status-hint rendering, and chat-to-chrome invalidation.
• Observed that two appended blocks while scrolled up produce an unread count of 2; returning to offset 0 clears it; a later scroll-up starts from 0; a new arrival increments it to 1; and follow clears it.
• Observed the status hint '↓ 2 new · esc follow · pgup/pgdn scroll', and noted that Dirty::mark_chat marks chat, tail, and chrome.
• Confirmed the draw path contains neither a draw-time mark_chrome() call nor unread-specific chrome promotion.
• Ran the full current TUI test suite: 248 tests passed, 6 were ignored, and none failed.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "fix(tui): drop the draw-time chrome prom..." | Re-trigger Greptile}