[omartelo]

Summary

Hiding the pull request list in Pulls did nothing until something else asked for a frame. The hide button's click was handled in listHeader, after pullsScreen.layout had already read listHidden to place the slide, and nothing asked for the next frame. Measured on a rig: 3 out of 3 clicks left the list on screen; one slide began 6.1s later, on the next GitHub recheck. Show worked by luck: the frame after its press happened to read the new state.

Both clicks (toggleList in the list header, showList in the detail toolbar) are now handled at the top of pullsScreen.layout, before p.slide.at reads listHidden, the way the dock's act handles its header. Same family as #905 and #908.

Test

TestPullsListSlidesFromTheFrameThatTookTheClick clicks hide, then show, through the frame rig. The rig's usual click check does not catch this: a slide's first frame draws where it stood, so the frame that took the click and the next one match. The test asserts what the window needs instead, that the frame that took the click asks for the next one. It fails without the fix and passed 50 runs with it.

Test plan

• gofmt, go vet ./..., go test ./... in native/
• New test fails before the fix, passes after (-count=50)
• Rig on its own backend in a virtual display with a per-frame trace, to measure the bug before the fix