probeContent

Settle GeoTiffContent.AMBIGUOUS from the pixels, returning contentHint unchanged for files the tags already decided.

Reads a few blocks of the coarsest overview — cheap, but it is file I/O, so call it off the render thread. Two observations are conclusive: a negative sample can only be a height, and a sample beyond TERRAIN_CEILING_METRES can only be a sensor count, which is what separates full-range 16-bit imagery (Landsat, Sentinel) from terrain. A raster whose values fit both readings stays GeoTiffContent.AMBIGUOUS rather than being guessed at: 12-bit imagery and lowland terrain occupy the same numbers, and nothing in the pixels distinguishes them.