cq: surface thinking-block content from messages (currently dropped in array_msgs)

claude_messages_sql's array_msgs CTE in src/views.rs (~lines 123-126) picks exactly one content block per message to extract into messages.text:

(SELECT json_extract_string(item, '$.text')
 FROM (SELECT UNNEST(CAST(json_extract(json, '$.message.content') AS JSON[])) AS item)
 WHERE json_extract_string(item, '$.type') = 'text'
 LIMIT 1) AS text,

The WHERE item.type = 'text' filter means thinking blocks never survive into messages, or anything downstream of it (trace, search, etc.). This isn't the empty-signature "encrypted thinking" case from some fixtures. A live grep of real transcripts under ~/.claude/projects/*/*.jsonl shows thinking content that's often substantial and non-trivial, e.g. "I should look up Jira issue AIDEV-509 to find its associated epic...". There's real content being silently dropped.

Design questions this issue should resolve before implementation, not after:

1. Storage shape: a new thinking_text column alongside messages.text, or a separate view/CTE? A single message can carry both a thinking block and a text block, so array_msgs can't just swap which type it filters for.
2. Scope: does this belong on messages at all, or does tool_calls/the trace Span/Gap model in src/trace/mod.rs also want thinking content? If messages is the only consumer, this is a cq-core feature, not a trace one.
3. If thinking content does need to reach the trace emitter eventually, the existing args.detail mechanism (from #46/#47, in src/trace/perfetto.rs) should cover it once the data exists in messages, no new mechanism needed.
4. Size/redaction: thinking text can be long, and occasionally sensitive in the same way tool args can be. Not investigated yet.

Scoping only. No implementation in this issue.

Start in src/views.rs at the array_msgs CTE and trace its consumers, then inspect src/trace/mod.rs and src/trace/perfetto.rs for existing thinking-content paths. Compare the available storage options, trace scope, and size/redaction concerns using real transcript examples. Done means the storage and scope decisions are documented clearly enough to define a follow-up implementation.