fix(renderer): avoid blanking popups before painting content

[jiunbae]

Problem and change

An updating display-popup currently paints a rectangle of spaces before repainting its content. Even a tiny TUI update therefore emits a complete blank popup followed by the actual rows. Terminals consuming the output incrementally (including over PTYs/SSH) can show that intermediate blank surface as a flash.

Render each popup content row directly in its final state instead. Surface rows are padded after their content, using the popup style and the existing ANSI/grapheme-aware truncator, so shorter/missing rows still erase stale cells. Static formatted text uses its existing fixed-width renderer for every row. Borders and process SGR remain supported.

Related to #216, but this does not fix its per-read-chunk refresh/backpressure issue or introduce incremental/delta popup frames. Host status ticks overwriting an open popup are reported separately in #225.

Validation

• Regression tests check that a filled borderless popup has no all-blank prepaint and each row is positioned once.

• A screen-model test verifies that short Unicode/combining-character content and missing rows clear old text.

• Existing overlay tests cover SGR, literal format-like process output, clipping, row bounds, border variants, and static text.

• Linux x86_64, isolated 120x30 PTY clients, status disabled: six seconds of the same idle muxa watch popup emitted 57,624 bytes / 180 full-width blank runs with stock rmux 0.10.0, versus 32,298 bytes / 0 full-width blank runs with this patch. This measures removal of the blank phase; it is not a claim that all sources of terminal flicker or overlay backlog are fixed.

• cargo build --workspace: passed.

• cargo clippy --workspace --all-targets -- -D warnings: passed.

• cargo fmt --all -- --check: passed.

• cargo test -p rmux-server --lib renderer::overlay: 16 passed.

• cargo test --workspace --all-targets: stopped at the unchanged rmux_manpage_renders_with_system_formatter test. This Linux environment has Ubuntu's minimized-system /usr/bin/man stub, which returns success with a notice instead of rendering the page; the test consequently cannot find RMUX. No changes were made to that test or the manpage.

• The broader run SHELL=/bin/sh cargo test --workspace --all-targets --no-fail-fast -- --skip rmux_manpage_renders_with_system_formatter completed but was not green. Besides shell-name expectations and an unselected local Ruby version, it found a changelog release-heading violation (fixed in the follow-up commit), an attach redraw failure, and a server test process stack overflow.

• With SHELL=/bin/bash RBENV_VERSION=3.1.1, all four affected CLI/release test targets passed on rerun: cli_attach_flow (40), cli_window_surface (134), release_downstream_writer_spec (6), and release_gate_python_floor_spec (3).

• The server library rerun with SHELL=/bin/bash RBENV_VERSION=3.1.1 RUST_MIN_STACK=16777216 cargo test -p rmux-server --lib completed: 4,075 passed, 5 failed, 3 ignored. The failures are timed_out_attach_generation_cannot_consume_healthy_fallback_queries, display_message_expiry_preserves_an_open_bracketed_paste_boundary, managed_client_actions_fail_closed_when_a_pid_is_reregistered, rename_session_serializes_timer_rekey_before_source_name_reuse, and live_attach_clock_mode_reroutes_complete_bracketed_paste. These failures have not been compared against an unmodified upstream checkout, so this PR does not assert they are baseline failures.

This PR remains draft because whole-workspace validation is not green, although the popup regression tests and the isolated PTY comparison pass. The user's running rmux daemon was not replaced or restarted.