fix(tui): wrap ask_user transcript trace so it can't clip chat history

[Fullstop000]

Problem

While the agent is running, the status band at the bottom appeared to cover the newest chat-stream history — the freshest transcript lines vanished at the boundary with the status bar.

Root cause

render_transcript slices the transcript one logical Line per visible row and renders with Wrap { trim: false }. That contract holds only because block_lines pre-wraps every block to the terminal width via wrap_line.

The ask_user branch was the one exception: it did lines.extend(ask_user_resume_trace(entry)) and returned without wrap_line. ask_user_resume_trace packs the full untruncated question and answer into a single Line, so a normal one-sentence question produces a line far wider than the terminal. That one logical line re-wraps into several rows at render time while still being counted as a single row — the slice then renders taller than the area, ratatui clips the bottom, and the newest lines disappear behind the band.

Measured: that trace line was 156 cols on a 40-col terminal.

Fix

Route the ask_user trace through wrap_line like every other block type, restoring the "every committed line ≤ width" invariant. Bonus: wrapped continuation rows are now correctly indented instead of ragged at column 0.

Tests

• ask_user_trace_lines_fit_width — pins the width invariant.
• long_ask_user_trace_keeps_newest_line_visible — drives the real block_lines → transcript → render_transcript path on a TestBackend and asserts the newest line isn't clipped.

Both verified to fail without the fix and pass with it.

Dogfood

Real-binary /resume comparison (old vs new, cols=72 rows=12): the old binary clips the newest line behind the status band; the fixed binary shows it. Screenshots in a comment below.

Gates

cargo fmt --check ✓ · cargo clippy -D warnings ✓ · cargo test --workspace (588) ✓ · cargo build --release ✓

[Fullstop000]

Dogfood — real binary, before/after

Built the pre-fix (origin/master) and fixed binaries, crafted a session with a long ask_user question, and resumed it via /resume at cols=72 rows=12 (transcript tighter than content, so auto-follow pins to the bottom). Captured the actual terminal grid with pyte.

Notice (the newest line) visibility across sizes (cols=72):

rows	OLD (pre-fix)	NEW (fixed)
9	shown	shown
10	CLIPPED	shown
11	CLIPPED	shown
12	CLIPPED	shown
13	CLIPPED	shown
14	CLIPPED	shown

OLD (pre-fix) — newest "Resumed session" line missing (covered by band); wrapped rows ragged at col 0:

  ✓ ask_user · PR #127 green. Two feat: commits since v0.34.0. Add to
[Unreleased], or bump to v0.35.0 and cut a release?: Bump to v0.35.0 and
cut a release now

  ZZZ_NEWEST_REPLY: cutting v0.35.0
                                          <- "Resumed session" notice clipped here
┌ > ───────────────────────────────────────────────────────────────────┐

NEW (fixed) — newest line visible; continuation rows indented under the marker:

  ✓ ask_user · PR #127 green. Two feat: commits since v0.34.0. Add to
  [Unreleased], or bump to v0.35.0 and cut a release?: Bump to v0.35.0
  and cut a release now

  ZZZ_NEWEST_REPLY: cutting v0.35.0

  Resumed session  session-…-shipverify .
┌ > ───────────────────────────────────────────────────────────────────┐

[codecov-commenter]

Codecov Report

❌ Patch coverage is 98.11321% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 76.94%. Comparing base (126dcfa) to head (e35bd9a).

Files with missing lines	Patch %	Lines
ignis/src/console/render/blocks.rs	97.05%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #130      +/-   ##
==========================================
+ Coverage   76.62%   76.94%   +0.31%     
==========================================
  Files          66       66              
  Lines       17575    17671      +96     
==========================================
+ Hits        13467    13597     +130     
+ Misses       4108     4074      -34     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Fullstop000]

Follow-up: folded in the CJK / space-less wrap fix

Review flagged (and I confirmed empirically) that wrap_line only broke on ASCII spaces, so a space-less run wider than a row — CJK text, a long URL — stayed over-width and still clipped the newest transcript line. Since that's the bilingual ask_user case, I folded the fix into this PR: wrap_line now hard-breaks an over-width word at the display-column boundary, so the "every committed line ≤ width" invariant holds for all block types.

Verified on the real binary (/resume, pure space-less Chinese question, cols=72 rows=12) — before this commit the newest line was clipped and continuation rows were ragged at column 0; after, continuation rows are indented and the newest line shows:

  否能够在没有空格的情况下正确地按显示宽度断行否则最新的聊天记录就会被底
  部状态栏遮挡住:
  升级到下一个版本并且立即发布一个全新的正式版本不要再等待了

  最新回复哨兵SENTINEL

  Resumed session  session-…-cjk2 .
┌ > ───────────────────────────────────────────────────────────────────┐

New test: wrap_line_hard_breaks_spaceless_run. Per your call, holding under [Unreleased] (no version bump).