fix(tui): render thinking preview markdown and stabilize activity rows (#200)

* fix(tui): render thinking preview markdown

* fix(tui): bound thinking-preview comment regex

* fix(tui): keep activity rows visually stable

* docs(tui): changelog and design for clean activity rendering

* perf(tui): cache rendered thinking-preview markdown across live ticks

The markdown render + regex parse for the thinking preview can run many times
per second on Live refresh ticks when no new content has arrived. Introduce
a cache key (the preview string) so markdown parsing only runs when the
pending thinking text actually changes, not on every spinner animation.

Includes tests verifying cache behavior and output equivalence.

* test(tui): freeze clock in thinking-preview cache equivalence test

The output-equivalence test compared full Console.export_text() across two
compose() calls, whose status line embeds elapsed time and token rate. Freeze
time.monotonic so the two renders are deterministic and only the cached preview
governs the comparison; guards against a rare timing-boundary flake.

Author: elkaix

CHANGELOG.md

@@ -15,6 +15,10 @@ GitHub Releases page; `0.8.0` is the new starting line.
 
 ## Unreleased
 
+- **Thinking and subagent activity now render cleanly in the terminal.** Live reasoning previews
+  render complete Markdown without exposing top-level HTML comments, activity-tree rows remain
+  visually stable, and the coral shimmer is reserved for the active verb spinner.
+
 ## 0.57.0 (2026-07-05)
 
 - **No more ghost/duplicate input prompt while the agent works.** After

docs/en/release-notes/changelog.md

@@ -17,6 +17,10 @@ GitHub Releases page; `0.8.0` is the new starting line.
 
 ## Unreleased
 
+- **Thinking and subagent activity now render cleanly in the terminal.** Live reasoning previews
+  render complete Markdown without exposing top-level HTML comments, activity-tree rows remain
+  visually stable, and the coral shimmer is reserved for the active verb spinner.
+
 ## 0.57.0 (2026-07-05)
 
 - **No more ghost/duplicate input prompt while the agent works.** After