v1.79.0

Added

• Rerun visualization backend with seamless backend switching between holoviews and rerun (#755)
• ResultRerun type for dedicated .rrd result handling (#882)
• extra_panels parameter in to_auto_plots() for composability (#846)
• LEVEL_SAMPLES constant and BenchRunCfg.level_to_samples() for transparent level-to-sample lookups (#834)
• samples_per_var parameter on BenchRunCfg — explicit sample count that overrides level (#834)
• Improved BenchRunCfg docstring with quick-start examples and level-to-samples table (#834)
• Mermaid architecture diagram in concepts docs (#852)

Changed

• cache_samples is now opt-in, with auto-enable for progressive runs (#889)
• Refactored holoviews backend: unified tap logic, time HoloMaps, and filter usage (#754)
• All generated example filenames now prefixed with example_ (#890)
• Improved error messages, result validation, and example modernization (#895)
• Improved validation, error messages, and onboarding docs (#887)

Fixed

• Type hints and error handling for extra_panels (#896)
• Use random port instead of port=0 to avoid EADDRINUSE on Linux 6.x (#894)
• Save report synchronously before serving to prevent Bokeh race condition (#893)
• Coerce sweep bounds/values to declared type (#888)
• Use CDN viewer for rerun and eagerly init recording (#884)

Performance

• as_completed() for parallel result streaming (#891)
• Hoist shared allocations and pre-cache numpy arrays (#886)
• Bypass xarray indexing and cache get_input_and_results (#885)
• Batch cache lookups to reduce SQLite round-trips (#883)
• Memoize to_dataset() results (#877)
• Speed up RTD docs build (#892)