fix: guard empty depth maps during dense fusion

[woodongwong]

No description provided.

[cdcseacave]

Thanks for the report: fusion could indeed read an empty depth-map. The cause is different from eviction, though, so this is fixed at the source in #1309 instead of with a check in the fusion walk.

• Why eviction can't be the cause: DMapCache is an LRU cache, and eviction stops at the current reference (SkipMemoryCheckIdxImage). The reference is loaded before its neighbours, so every neighbour flagged for it is newer than the reference and stays cached while it is fused.
• The real path: a .dmap that fails to load (corrupt or truncated). DMapCache::UseImage ignored Load()'s failure and cached the empty map, and the fusion loops only asserted that the reference was non-empty.
• The fix in densify: T&T-tuned fusion, decoupled confidence recalibration, bounded fusion memory #1309:
  • UseImage logs the failure and doesn't cache the empty map.
  • Both fusion loops skip such a reference.
  • EstimateNormalMaps skips the map instead of aborting, which used to leave later neighbours without normals.

With that, a map that can't be read is never fused or walked into, and the fusion loop needs no per-probe branch. Closing in favour of #1309.