Make QuadSurface gen safe for parallel rendering

[LubuSeb]

Fixes #1046.

Summary

• make QuadSurface::gen() safe when vc_render_tifxyz calls it from parallel tile workers
• move the large coordinate/validity/normal scratch buffers from shared per-surface members to per-thread scratch buffers
• guard lazy normal-cache rebuilds and snapshot the cache header before warping normals
• add an empty-source guard to the nearest-neighbor warp helpers
• add a concurrent QuadSurface::gen() regression test that requests normals from multiple worker threads

Reproduction / Root Cause

Issue #1046 reports SIGSEGVs in vc-render-tifxyz after updating from an April build. The posted stack trace lands in:

QuadSurface::gen -> warpNearestConstVec3f -> orow[dx] = ... srow[sx_i]

vc_render_tifxyz renders tile columns inside an OpenMP loop, and each tile calls genTile(), which calls QuadSurface::gen() on the same QuadSurface.

The recent optimized gen() path reused mutable destination scratch buffers on the surface object:

• _genCoordsScratch
• _genValidScratch
• _genNormalsScratch

Those are written by each gen() call, so parallel tile workers could write and resize the same cv::Mat storage concurrently. The lazy _normalCache build was also unguarded, so first-use normal generation could race.

Scope / Minimality

This patch is limited to:

• volume-cartographer/core/include/vc/core/util/QuadSurface.hpp
• volume-cartographer/core/src/QuadSurface.cpp
• volume-cartographer/core/test/test_quadsurface_components.cpp

It does not change surface-normal orientation policy, vc_render_tifxyz command-line options, renderer output semantics, or scheduling.

Issue / Artifact Binding

• Issue: vc-render-tifxyz fails with sigsegv #1046
• Artifact: this pull request from LubuSeb:codex/villa-1046-threadsafe-gen
• Prize context: Vesuvius Challenge monthly progress prizes may consider VC/tooling improvements; this PR is the reviewable technical artifact, not a guaranteed award request.

Validation

Run in the project Docker builder:

cmake --preset ci-tests-gcc
cmake --build --preset ci-tests-gcc --target vc_render_tifxyz
cmake --build --preset ci-tests-gcc --target test_quadsurface_components
./build/ci-tests-gcc/bin/test_quadsurface_components
cmake --build --preset ci-tests-gcc
ctest --preset ci-tests-gcc --output-on-failure

Results:

• vc_render_tifxyz built successfully.
• test_quadsurface_components passed, including the new concurrent QuadSurface::gen() regression.
• Full configured CTest passed: 97/97 tests.
• git diff --check passed; only Windows LF-to-CRLF warnings were emitted by Git.

Known Limits

I did not have the private/large segment dataset that produced the original backtrace, so I validated the failure mode with a focused concurrency regression and the full Dockerized test suite rather than a direct data replay.

This is also a normal GitHub PR contribution. It may be relevant for Vesuvius monthly progress-prize review, but this PR does not claim or guarantee an award.

[vercel]

@LubuSeb is attempting to deploy a commit to the scroll Team on Vercel.

A member of the Team first needs to authorize it.