feat(tui): add --show-full-output and make Ctrl+O reveal untruncated output

[PerryLink]

Closes #730.

What

Long tool results and terminal-command output are previewed ([Preview only: …], src/tau_coding/tui/state.py:924) so one result cannot flood the transcript — but the hidden remainder was previously unreachable without exporting the session.

This adds both things the report suggested:

• --show-full-output — start the TUI with expansion on (per-invocation, not durable)
• Ctrl+O — the expansion toggle that already existed for tool results now also renders the untruncated text

How

Full text is retained per item (ChatItem.tool_result_full_text) and only when the preview was actually lossy, so memory does not grow for short results. Rendering goes through ChatItem.result_text(expanded=…), and the flag threads cli.py → run_tui_app → TauTuiApp → TuiState.

Preview defaults are unchanged — nothing expands unless asked for. The !-command path (app.py:5840, capped at 120 lines / 2000 chars) now retains its untruncated output too.

Files: 8 (+432/−20), including tests/test_tui_full_output.py (14 tests) and docs under website/content/.

Verification

• uv run pytest tests/test_tui_full_output.py → 14 passed
• Controlled comparison across every touched test file (test_tui_full_output, test_cli, test_tui_app, test_tui_components, test_tui_config, test_rendering, test_tool_history, test_tui_adapter) → 628 passed, 17 failed
• Those 17 are exactly the set measured with the change stashed (git stash, same files): Windows-only sidebar path/editor issues, drive-letter casing, and % formatting in print-mode prompts. Zero regressions.
• uv run ruff check . → All checks passed · uv run ruff format --check . → 196 files already formatted
• uv run mypy → 14 errors, all pre-existing fcntl.flock on Windows, none in the changed files

One caveat on methodology: the full suite is resource-flaky in this environment (52 vs 29 failures across two identical runs, killed early), so the no-regression claim rests on the controlled stash-vs-changed comparison above rather than on a full-suite count.