Smaller findings from a test run on moli 1.1.15 #1291

1. networkidle timeout is silent. moli fetch --wait-until networkidle --dump markdown https://www.nrk.no takes the full 20s timeout and prints no warning, although the help text says it returns "with a warning".
2. Full-page screenshot limits. --dump screenshot_full of https://no.wikipedia.org/wiki/Oslo fails with 815865700 RGBA bytes, exceeding the 268435456-byte transient budget. Playwright screenshot({ fullPage: true }) fails with raster dimensions 1920x100165 exceed backend limit 65535. Tiling or downscaling, or at least a documented hint in the error, would help.
3. Emulation.setDeviceMetricsOverride with mobile: true returns does not support mobile=true. This blocks Playwright and Puppeteer device presets.
4. The End key does not scroll. Input.dispatchKeyEvent keyDown and keyUp End (keyCode 35) leaves scrollY unchanged on Wikipedia, NRK and MDN, while mouseWheel and scrollTo work.
5. Text overlap on VG article pages. On a vg.no article the headline overlaps the lead paragraph in a --layout screenshot. This may be related to issue 2.

What worked well, for context: 30 parallel fetches across 25 sites in 20s. 20 of 21 raw CDP interaction tests passed, including clicking inside a cross-origin consent iframe. The Playwright and Puppeteer basics work (goto, click, fill, wheel, screenshot, pdf, contexts, console, response events). The bot.sannysoft.com checks pass except for broken-image dimensions. Memory stayed stable at 130 to 150 MB RSS.