fix: cycle thinking level in lean tui

[rumpl]

Summary

• Handle Shift+Tab in the lean TUI to cycle the current agent's thinking-effort level
• Update the lean footer immediately and show a lightweight notice for success/unsupported/error cases
• Add lean TUI tests for cycling and unsupported models

Validation

• go test ./pkg/leantui
• go test ./pkg/app ./pkg/tui ./pkg/leantui
• go vet ./pkg/leantui
• go build ./...

Note: task lint could not be run because task is not installed in this environment. go test ./... was attempted and hit unrelated existing/environmental failures in pkg/cache and pkg/rag/treesitter (CGO_ENABLED=1 required).

[docker-agent]

Assessment: 🟢 APPROVE

Reviewed pkg/leantui/update.go and pkg/leantui/update_test.go.

The Shift+Tab → handleCycleThinkingLevel implementation is correct:

• Nil guard on m.app prevents panics before any app is attached.
• SupportsModelSwitching guard correctly surfaces a user-friendly notice for remote runtimes before attempting the cycle.
• Error branching handles both runtime.ErrUnsupported (model-specific, shown as muted notice) and generic errors (shown with error styling) without updating m.status.thinking on failure — correct behaviour.
• Status + notice update on success is appropriately minimal; the synchronous write to m.status.thinking followed by the async TeamInfoEvent from refreshAgentInfo is safe because both writes land on the lean TUI's single event-loop goroutine via the select in leantui.go.
• Tests cover the success path and the ErrUnsupported path with a well-structured mock.

No bugs found in the changed code.