Show real-time thinking summaries during extended thinking to distinguish from API hangs

Problem

When Claude Code enters extended thinking mode, the terminal only displays thinking... with no further output. During long thinking phases (3-5+ minutes), it's impossible for users to tell whether:

• The model is genuinely reasoning through a complex problem, or
• The Anthropic API is hanging/experiencing an outage (which happens regularly)

This creates uncertainty and wastes time — users end up waiting for a response that may never come, or canceling a request that was actually making progress.

Current behavior

• thinking... is shown as a static label with no content
• The redact-thinking API flag (introduced ~v2.1.68) strips thinking content server-side, so even with showThinkingSummaries: true and alwaysThinkingEnabled: true, there's nothing to display
• Ctrl+O toggles verbose mode but doesn't help since the thinking content itself is redacted
• Elapsed time is shown and counts up, but that alone doesn't confirm the model is still actively processing

Proposed solution

Stream thinking summaries in real-time during the thinking phase, similar to how response tokens are streamed. Even short periodic summaries (e.g., "Analyzing code structure...", "Evaluating approach...") would give users confidence that the model is still active and making progress.

Alternatively, at minimum show additional progress indicators (token count, heartbeat/activity signal) alongside the existing elapsed timer so users can distinguish an active thinking session from a stalled API connection.

Impact

This is a significant UX issue for users working with complex tasks where extended thinking is common. The uncertainty leads to unnecessary cancellations and erodes trust in the tool's reliability.