perf(ui): drop unconditional per-frame repaint — idle window goes to ~0% CPU

[weter11]

Summary

update() ended with an unconditional ctx.request_repaint(), forcing egui to render a full frame every vsync. On a 165 Hz display that pegs the UI thread even when the window sits untouched — the other half of the idle-CPU burn alongside #85's cover-fetch gate (this stacks on top of it).

What changed

The four channel drains in update() now return whether they drained anything; the unconditional repaint is replaced by a scoped policy:

Dependency (per-frame tick)	Resolution
image_rx drain (cover art)	repaint on drain; poster Spinner self-repaints while fetch pending
per-task download rx drain	repaint on drain + request_repaint_after(250ms) while tasks active (ProgressBar is passive; ETA is Instant-based and counts by wall-clock)
play_result_rx drain	repaint on drain + request_repaint_after(1s) while rx is Some (game exit / login-required can arrive while idle)
operation_rx drain (45 async ops)	repaint on drain
Spinners (account, proton list, steamguard, poster)	egui 0.33 Spinner self-requests repaint while visible — no change needed
TextEdit cursor blink / scroll inertia	egui internal — no change needed

With nothing pending, no repaint is scheduled and the main thread blocks in the event loop until the next input event (egui repaints on input itself).

Measured verification (idle window, not just compile)

Same binary pair, same Xvfb display, fresh config (no Steam session), 10 s samples:

Metric	before	after
UI thread, idle	353% of one core	0% of one core
whole process (top-style)	331%	1.6%
input wake (40 XTEST keys)	—	288% burst, settles back to 0%

The window was verified alive and input-responsive after the change (focus + synthetic input → repaint burst → returns to idle), not assumed.

Test Plan

• cargo check --all-targets clean
• cargo test --lib — 141 passed
• Live UI smoke test on real display (user)