Gracefully degrade QC plot rendering failures

[karl-koschutnig]

Summary

• Wraps the lpp.plot_tria_mesh(...) / fig.write_image(...) calls in qcPlots() (mesh, profile, and hull stages) in try/except, so a rendering backend failure logs a warning and skips that image instead of aborting the whole run.
• This addresses the "worse half" of QC plots fail in the 0.10.1 Docker image: "Kaleido requires Google Chrome to be installed" #24: a fully successful surface reconstruction is no longer discarded just because a QC PNG couldn't be rendered (e.g. kaleido 1.x requiring an external Chrome/Chromium binary). The kaleido/Chrome dependency question itself is left for separate follow-up, as discussed in the issue.

Ran ruff check and ruff format --check on the changed file per the maintainer's request.

Fixes #24

Test plan

• ruff check hipsta/utils/qc_plots.py passes
• ruff format --check hipsta/utils/qc_plots.py passes
• Maintainers to confirm on a pipeline run where kaleido/Chrome is missing that the run now completes with a logged warning instead of aborting

[m-reuter]

thanks for the contribution, looks good to me. Did you test this, e.g. you can map this file into the container and re-run? Would be great if you could do this, just to confirm we did not miss anything else. Thanks.

[karl-koschutnig]

Yes — tested this against the actual `deepmi/hipsta:0.10.1` image (`kaleido 1.3.0`, `plotly 6.8.0`) using the reproducer from #24, mounting the patched `qc_plots.py` over the installed one:

• Baseline (unpatched 0.10.1): reproduces the exact traceback from QC plots fail in the 0.10.1 Docker image: "Kaleido requires Google Chrome to be installed" #24 — crashes in `qcPlots()` at the mesh stage.
• Patched: logs `[WARNING: qc_plots.py] Could not create QC plot for mesh stage: ...` and continues into `checkSurface()` instead of aborting. That particular subject's segmentation then hit an unrelated, pre-existing `AssertionError` ("Surface contains holes") — confirmed this also happens on the unpatched baseline with `--no-qc`, so it's a property of that segmentation, not something the patch introduced.
• Ran a second subject through the full pipeline end-to-end (all three QC call sites — mesh/profile/hull) with the patch applied: all three log the same graceful warning, the `qc/` output directory ends up empty (as expected, since kaleido/Chrome still can't render), and the run finishes with `Hipsta finished without errors.`

So the try/except wrapping is confirmed working as intended — a rendering failure no longer discards a successful run, and I didn't see any other regressions during a full run.

Also opened #27 for the Docker Chrome install, since you mentioned being open to that as a separate stopgap — with that Dockerfile change, the same full pipeline run produces valid, non-empty QC PNGs at all three stages with no warnings, tested both as root and as a non-root `--user`.