A new gsParaview class to sanitize paraview file I/O - #876

src/gsIO/gsParaview.hpp:1

• openIfRequested always opens fn + ".pvd", but several write paths can legitimately produce only .vtu (singleFile + writePvd=false / skipPvd) or .vtp (curve/mesh). This makes the show option unreliable and can open a non-existent file. Consider tracking the actual produced “entry file” per write call (e.g., .pvd when written, otherwise the concrete output like .vtu/.vts/.vtp) and open that instead.
  src/gsIO/gsParaview.hpp:1
• The gsParaview option "writePvd" is explicitly not honored for the non-singleFile, non-bezier multipatch export path, which is surprising for API consumers and makes behavior inconsistent with write(geometry) / write(field). A tangible fix is to add a skipPvd (or writePvd) parameter to the relevant gsWriteParaview(const gsMultiPatch&, ...)/vector overloads (or route this path through the gsParaviewCollection object conditionally) so "writePvd" is consistently applied.
  src/gsIO/gsParaviewCollection.hpp:1
• Explicit timesteps are truncated to an integer (cast<T,int>(time)), which loses information for non-integer times (e.g., 0.1, 0.5) and will write incorrect timestep="..." attributes in the .pvd. Since the public API accepts T time, m_time should store the full T value (or double) without integer truncation.
  src/gsIO/gsParaviewDataSet.hpp:1
• isEmpty() always returning false makes the method misleading and undermines the defensive checks in gsParaviewCollection (!dataSet.isEmpty()). Even if an “empty dataset” is now structurally impossible (because constructors require a geometry reference), it’s better to reflect actual state (e.g., m_geometry == nullptr or m_filenames.empty()) or remove/replace isEmpty() from the contract to avoid future misuse.
  src/gsIO/gsWriteParaview.hpp:1
• After detecting !file.is_open(), the function continues writing to file unconditionally, which will silently drop output and can mask failures. This should return early (or throw/assert) once opening fails, especially since the function later calls makeCollection, potentially creating a .pvd that references a missing .vtu.
  src/gsIO/gsWriteParaview.hpp:1
• VTK inline binary (base64) expects the 64-bit byte-count header to match the declared byte_order (and typically assumes little-endian when byte_order="LittleEndian"). Here the header is emitted in host endianness with no conversion, so output can be invalid on big-endian systems. Consider explicitly serializing the header (and potentially numeric payload) to the chosen byte order before base64 encoding, or enforce/validate little-endian when export_base64=true.
  src/gsIO/gsWriteParaview_.cpp:1
• The Python binding renames the keyword argument from controlNet to ctrlNet, which is a breaking change for users relying on keyword args. To maintain compatibility, keep the existing keyword (controlNet) (even if the C++ parameter is ctrlNet) or provide both names (e.g., keep controlNet in py::arg).
  src/gsAssembler/gsExprEvaluator.h:1
• This “backward compatibility” logic will prefer numPoints/plotElements whenever those keys exist, even if users set only the legacy keys (plot.npts / plot.elements). Since defaultOptions() now adds the canonical keys, exists("numPoints") will likely always be true, effectively ignoring legacy values. Align this with the more robust alias logic used in gsParaviewDataSet::getNumPoints/getPlotElements (e.g., prefer canonical only when it differs from its default, otherwise fall back to legacy if it differs from default).
  unittests/gsParaview_test.cpp:1
• The test likely doesn’t exercise the precision behavior because it guesses output filenames as fn + "0.vts" (no delimiter), while other tests already acknowledge multiple naming schemes (e.g., _0, _patch0, etc.). As written, the check can silently skip on most outputs, weakening coverage. Consider resolving the actual produced .vts filename(s) (e.g., using the configured patch delimiter, or searching expected variants and asserting one is found) and then asserting on size/content.