Clearer chat, current models, and steadier sidepanel work

This release makes day-to-day sidepanel work feel steadier. Provider failures such as running out of credits now show a clear explanation instead of a generic stream error. Thinking models surface their steps as collapsible headings while they work. Chats no longer bleed into each other if you switch tabs mid-reply. The model list is trimmed to current options, and if a saved model has been sunset SupaPM switches you to its successor and asks you to confirm.

You can also see token usage per reply and for the whole session, and check how much browser storage SupaPM is using in Settings.

What's new

• Provider errors unpack billing, API key, rate-limit, and context-window failures into actionable messages, with an Open settings shortcut when relevant.
• Failed replies appear as error cards and stay out of the next model request so they do not pollute conversation context.
• Thinking-model reasoning streams as collapsible step headings; the active step shimmers until the next one arrives.
• Per-turn and session token usage is visible in chat, including thinking tokens when the provider reports them.
• Model picker shows current-generation options only (for example GPT-5.6 Sol / Terra / Luna). Sunset models remap to a successor with a confirmation dialog.
• Sidepanel toasts use clearer coloured alerts with icons and a dismiss control.
• Slash shortcuts only open when / stands alone (for example after a space), so mid-word typing does not pop the prompt menu.
• Chats stay isolated while generating: switch tabs or start a new chat without losing an in-flight reply.
• Settings shows local storage usage for SupaPM data in this browser profile.