test: input_decoder: workaround test failure by sending preliminary pictures

This MR workaround the deadlock from #27329 (closed), which is not what is tested here (especially since it happens pseudorandomly in the test).

Fixing this issue was only possible by workarounding another issue in the video output, where the current picture being used could be reset to NULL because of a flush. The issue itself has only been workarounded. A reliable test will come for this video output issue later.

Fixes #27329 (closed)