Research: MPRIS event fingerprint — what sidra actually emits

Part of #78

## Question

Produce the **authoritative fingerprint of MPRIS/playerctl events** for sidra (and, for contrast, chromium-throttled), so the SessionTracker stops misclassifying real signals:

Edits to / surprises in: `PlaybackStatus`, `Position` cadence (per-sec lines burst then 60s+ silence while Playing holds), `Seeked`, metadata updates (`xesam:title` empty-string lines), the transition theatre (`Playing ""` → `Stopped ""` → `Stopped ` → `Playing `, positions ≈1s), **deregistration blips from `playerctl -l`** (does sidra re-register between tracks?), and anything D-Bus/playerctl-wide (border case: multiplexed stream with chromium).

Evidence so far: 2026-08-09 live captures in docs/research (tracks table burst/silence, follow-stream dumps, pollwatch). Reproduce with the live follow stream + raw `dbus-monitor org.mpris` if needed.

Deliverable: produce `docs/research/mpris-events.md` (branch `research/mpris-events`), recording: for each event type — when it fires, what playerctl lines it shows, known quirks, and consequences for a session tracker. This unblocks the session-boundary grilling and the no-compression prototype.

Start by checking the existing evidence in docs/research (tracks table burst/silence, follow-stream dumps, pollwatch). On branch `research/mpris-events`, reproduce with the live follow stream plus `dbus-monitor org.mpris` and `playerctl -l` around track transitions, then compare sidra and chromium-throttled behavior. Create `docs/research/mpris-events.md` with event-by-event entries for when each event fires, the matching playerctl lines, quirks, and session-tracker consequences; done when the fingerprint and impact are fully documented.

Escrita pelo modelo de indexação a partir do texto da issue.