Memory graph and browser interaction improvements, clearer model settings

New features

• Improved domain browsing, cluster layouts, and motion in the memory graph, with smoother performance on large graphs.

Improvements

• Model settings show the reasoning levels supported by the selected model, including Off only when available. Your preferred level is retained when switching models.
• The maximum context compaction threshold now reflects the selected model's capacity, reserved output, and buffer, while retaining your saved threshold preference. Requires a device update.
• Custom API models support editable context capacity and maximum output, direct input for reasoning strength and compaction thresholds, and deletion of custom providers with their stored keys. Requires a device update.
• Improved browser waits for page loading, elements, and events, along with uploads and file selection. Added download management to view progress, pause, resume, cancel, and open downloaded files. Requires a device update.
• Unified icons for schedule statuses, tool calls, settings groups, and task goals. Goal icons reflect active, completed, and blocked states, while tool calls and failures are clearer in live schedule traces.

Fixes

• Schedule execution history renders results as Markdown, making code, lists, and links easier to read.
• Restored separate Plugins and Agents entries in Settings.
• Fixed browser pages being stopped when a tool's observation wait timed out. Slow pages can continue loading and be checked again later. Requires a device update.
• Fixed desktop clicks or input failing to reach some native system panels on macOS.
• Fixed macOS updates prompting for a restart before installation preparation finished. The app now waits for the native updater to be ready and provides clearer retry handling after failures.