feat(mpris): make the player rows tell you which is which

[Zakkaus]

Builds on #22. Review only feat(mpris): make the player rows tell you which is which.

Problem, from a real desktop

With Firefox and Chrome both playing, #22's picker lists four rows:

Chrome
Mozilla firefox
Mozilla Firefox
Google Chrome

Each browser publishes its own MPRIS endpoint, and Plasma Browser Integration bridges the same tab, so half the list is duplicates — and two of the names differ only in one letter's case. There is no way to tell which row is the music you want.

Fix: say what each endpoint is doing

自動選擇
Chrome · 播放中 · 下山 (梦幻西游方寸山门派曲)
Mozilla firefox · 播放中 · 大石碎胸口
自動 · Mozilla Firefox · 播放中 · 大石碎胸口 - 万能青年旅店
Google Chrome · 播放中 · 下山 (梦幻西游方寸山门派曲) - 要不要买菜

Identity, playback status, and the loaded track. Now the two Firefox rows are visibly the same song and the two Chrome rows are visibly the other one, so the choice is obvious. The row the automatic selection would pick is marked, so a user who does not want to choose can see the default is already sensible.

Metadata and PlaybackStatus ride on the D-Bus round trip that already fetched Identity, so this costs no extra traffic.

What it deliberately does not do

No deduplication. Two endpoints for one browser are a real choice, not noise. The bridge usually reports a cleaner artist — that is why the automatic ranking prefers it, as the comment above _selection_score explains — but "usually" is not "always", and hiding the wrong one would leave the user with only the broken option.

Two defects fixed while building it

Status was untranslated. MPRIS reports it as a protocol constant, so the first version showed Playing inside an otherwise Chinese row.

A delegate segfaulted the suite. Eliding long rows with a QStyledItemDelegate looks right and is not: initStyleOption runs during sizeHint, where the option's widget can already be destroyed, and Qt crashes inside QWidget::style(). The backtrace was QComboBox::changeEvent → QListView::doItemsLayout → QStyledItemDelegate::sizeHint → QWidget::style(). Eliding now happens where the row is built, which is our own string and needs no delegate.

Validation

• QT_QPA_PLATFORM=offscreen uv run pytest — 386 passed, run three times for the segfault
• uv run ruff check .
• uv run ty check

Checked against a live session with four real endpoints, rendered in zh-Hant and en — the rows above are the actual output, not test data.

Follow-up: the marker read as a second automatic option

Verified with two browsers publishing on the session bus at once. The rows come out as intended — Firefox · Playing · アイドル by YOASOBI against Chromium · Paused · Bad Apple!! by nomico — but the row automatic selection had landed on was prefixed with the word automatic, directly under the row labelled Automatic. Two rows opening with the same word is the confusion this picker exists to remove, so the marker now says which row is the current one instead.