module placement: the five concurrent fits each keep every thread

One Ceres fit keeps only ~5-6 of 32 cores busy whatever it is given, and on a loaded machine (load
28-36, JF16M int16, 4000 strong frames) the five side-by-side fits measured 10.9 s with every thread
each, 13.8 s with half, 14.6 s with a fifth, against 18.9 s one after the other (bb7c9c). The
thread split is dropped: each fit gets gr_settings' threads, as before the fits ran side by side.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015CeEGgpKdfkCTJb3YNMJcu

rugnux/RugnuxStills.cpp

@@ -340,14 +340,13 @@ void Rugnux::RefineStillsGeometry(int start_image, int end_image, int images_to_
 
         // The four held-out fits and the placement on every frame do not depend on each other, and one fit
         // keeps only five or six cores busy however many threads it is given (measured on 32): all five run
-        // side by side, each on a fifth of the threads. The placement is computed whether or not the held-out
-        // residuals then call for it.
+        // side by side, each with every thread - on a busy machine that measured faster than a fifth each
+        // (JF16M: 10.9 s against 14.6 s, 18.9 s one after the other). The placement is computed whether or
+        // not the held-out residuals then call for it.
         TimingMark placement_mark(logger, "module placement fits");
-        const int fit_threads = std::max(1, config_.nthreads / 5);
         GeometryRefinerSettings st = gr_settings;
         st.modules = start;
         st.refine_modules = true;
-        st.num_threads = fit_threads;
         auto placement = std::async(std::launch::async, [&] {
             return RefineGlobalGeometry(nominal, *cell, placing, st);
         });
@@ -357,7 +356,6 @@ void Rugnux::RefineStillsGeometry(int start_image, int end_image, int images_to_
             GeometryRefinerSettings sf = gr_settings;
             sf.modules = modules;
             sf.refine_modules = refine;
-            sf.num_threads = fit_threads;
             const auto f = RefineGlobalGeometry(nominal, *cell, half[fit], sf);
             if (!f.ok)
                 return std::numeric_limits<double>::infinity();
@@ -366,7 +364,6 @@ void Rugnux::RefineStillsGeometry(int start_image, int end_image, int images_to_
             GeometryRefinerSettings ev = gr_settings;
             ev.refine_beam_distance = false;
             ev.modules = f.modules;
-            ev.num_threads = fit_threads;
             const auto e = RefineGlobalGeometry(g, f.cell, half[1 - fit], ev);
             if (!e.ok)
                 return std::numeric_limits<double>::infinity();
@@ -395,7 +392,6 @@ void Rugnux::RefineStillsGeometry(int start_image, int end_image, int images_to_
                 // so every spot found and every prediction integrated from here on uses it. The image is
                 // not touched.
                 if (config_.module_diagnostics) {
-                    st.num_threads = gr_settings.num_threads;   // its fits run one at a time
                     ModuleDiagnosticsInput diag{nominal, *cell, placing, centres, start, st, rm,
                                                 experiment_.GetDetectorSetup().GetSensorMaterial(),
                                                 experiment_.GetDetectorSetup().GetSensorThickness_um()};