Method frame_end_cpu

native/shared/src/profiler.rs:275–298

CPU-only end-of-frame: histogram update + drain into rolling. Split out so tests don't need a wgpu::Device. Production callers go through `frame_end` which handles GPU readback first and then delegates here.

(&mut self)

    /// callers go through `frame_end` which handles GPU readback
    /// first and then delegates here.
    fn frame_end_cpu(&mut self) {
        // Phase 8 — sum the per-pass samples into a per-frame total
        // for the histogram before draining. CPU sums every sample;
        // GPU sums only those with timestamps (rest have None).
        let mut frame_cpu = 0.0;
        let mut frame_gpu = 0.0;
        for s in &self.frame {
            frame_cpu += s.cpu_us;
            if let Some(g) = s.gpu_us { frame_gpu += g; }
        }
        self.frame_total_cpu_us[self.histogram_idx] = frame_cpu;
        self.frame_total_gpu_us[self.histogram_idx] = frame_gpu;
        self.histogram_idx = (self.histogram_idx + 1) % ROLLING_FRAMES;
        self.histogram_filled = (self.histogram_filled + 1).min(ROLLING_FRAMES);

        for s in self.frame.drain(..) {
            let entry = self.rolling.entry(s.label).or_insert_with(RollingStats::new);
            entry.push(s.cpu_us, s.gpu_us);
        }
        self.open_cpu.clear();
        self.next_query = 0;
        self.pending_gpu.clear();
        self.frame_count = self.frame_count.wrapping_add(1);
    }

    /// Phase 8 — frame-history snapshot for the overlay's bar chart.
    /// Returns `(cpu_us, gpu_us)` pairs in chronological order, oldest