Feature request: opt-in thinking block persistence in session transcripts

Since approximately v2.1.76, thinking blocks in session JSONL transcripts have thinking: "" (empty content) with only the signature field preserved. Prior to this version, thinking content was stored in full.

Current behavior

{"type": "thinking", "thinking": "", "signature": "EsUCCkYICxgCKkD9icI6..."}

The thinking field is stripped before writing to disk. The signature is protobuf metadata — it does not contain or encrypt the thinking content. Not recoverable.

Requested behavior

An opt-in config option in settings.json:

{"persistThinkingBlocks": true}

When enabled, full thinking text written to local transcript alongside the existing signature field.

Why this matters

Session transcripts are user-owned data stored on the user's own machine. Thinking blocks represent the model's analytical reasoning and are valuable for:

• Session recovery: reconstructing analytical context after crashes or compaction
• Debugging: understanding why the model made specific decisions
• Multi-session research continuity: long-running projects benefit from reviewing prior reasoning, not just outputs

The infrastructure already exists — the content is generated and transmitted, the stripping happens at the persistence layer. Giving users control over local storage of their own data is low-risk and high-value. Should be opt-in, not default.