Dense: Fix out of memory edge case due to footprint explosion

[pierotofy]

Hey 👋

I've tracked down a recurring (but infrequent) issue that would cause OpenMVS to attempt to resize a depthmap to a very large size and causing:

Estimated depth-maps 134 (87.58%, 3m14s, ETA 27s)...terminate called after throwing an instance of 'cv::Exception'
what():  OpenCV(4.12.0) /code/SuperBuild/src/opencv/modules/core/src/alloc.cpp:73: error: (-4:Insufficient memory) Failed to allocate 68896563262380 bytes in function 'OutOfMemoryError'

This seems to happen when noise from an input sparse point cloud places a point really, really close to a view. When that happens, the footprint1 can end up being (focal / very very small depth value) --> very large number. In turn, this causes fScaleRatio to be set to a very large number, which pushes the avgScale value, and thus neighbor.scale to be a very large number.

This causes the call to:

if (DepthData::ViewData::ScaleImage(viewTrg.image, viewTrg.image, viewTrg.scale)) // <-- viewTrg.scale could be really large

In InitViews to crash the program.

One possible solution here is to bump the depth checks from 0 to a reasonable epsilon (I've picked 0.01, because the dataset where I reproduced this had a depth value of 0.000001), or another one could be to track a median scale rather than average (which would filter out noisy scale values).