By: cedric <cedric@sigrist.ch>
Bound sampler generations through live queue tuning Default max_generation_size to 262144 and apply changes to the next draw, preserving buffered work, task budgets, and generation feedback boundaries. Simplify native and Python samplers around generate(max_samples), update the v4 process contract and dashboard, and cover live tuning and recovery. Add the constant-batch-time amortization sweep, align benchmark generation limits, and record the generation-size experiments and updated plots/data. Make frontier planning tests independent of host CPU count and cover the minimum-core boundary, including explicit CPU limits.