Inferring 'value is 0/1' from history breaks across capture gaps

A coarse waveform renderer drew a flat level line in a bucket with no change only when an earlier change bucket in the same tile had no x/z values, on the logic that no change since then means the value is still 0/1. A recording with capture gaps breaks this. The resume value after a gap is written as a keyframe, but it is deliberately not counted as a change, so a resume to x or z left the 'known 0/1' state set, and the renderer drew a 0 line (the value bit of x) where the exact renderer drew x. Reset any carried-over inference at every discontinuity the data source can have (gap ends, file-segment resets), and test with a resume value that differs in kind (x/z) from the value before the gap.