Skip a flush that races between resize() and render()

[afonsojanu]

Fixes #1780

resize() updates the renderer's width and height and arms a redraw, but it never touches the stored view — that only happens afterward, when render() runs with a fresh View laid out for the new size. Both calls happen back to back in the same event loop iteration, but the ticker goroutine calls flush() independently at 60Hz, and if a tick lands in the gap between them, flush() sees the new size paired with the previous size's view. It draws that stale content into a buffer already resized for the new dimensions, and the next tick then corrects it a frame later — which is exactly the "ghost frame" the issue describes and traces in detail.

flush() now checks a viewStale flag that resize() sets and render() clears, and returns immediately without drawing while it's set. The pending erase and the width/height change stay in place, so once render() catches up (at most one tick later, since it always follows resize() in the same event loop iteration), the very next flush() draws the correct view at the correct size in one pass instead of two.

Added a direct test on the renderer driving the exact sequence the race produces: resize, flush before render, render, flush again — checking that nothing reaches the output writer during the gap and that the cell buffer only resizes once the correct view is available. Confirmed via git stash that this test fails (drawing the stale "old" view) without the fix and passes with it.

• go test ./...: all passing.
• go test -race ./... on the affected tests: clean.
• go vet ./... and gofmt -l: clean.
• golangci-lint run ./...: 0 issues.