fix: preserve line breaks in thinking output

[Huarong]

Summary

This change preserves line breaks in text-mode thinking output instead of collapsing all whitespace into a single line.

Problem

agent_thought_chunk content is currently rendered through collapseWhitespace(...), which flattens \n into spaces before printing. This makes multi-line reasoning hard to read and loses structure that agents intentionally send.

Changes

• preserve \n when flushing buffered thought text
• normalize CRLF to LF before rendering
• render multi-line thought output as a [thinking] block instead of a single flattened line
• add a regression test covering blank lines and line breaks in thought chunks

Why this should be separate

This PR only fixes formatting/readability. It does not change when thought chunks are emitted.

Testing

• ./node_modules/.bin/tsc -p tsconfig.test.json
• node --test dist-test/test/output.test.js

[dutifulbob]

Triage result

Human attention: ⚠️ Required
Recommendation: 🏁 escalate to a human
Human decision needed: ready for human landing decision

Quick read

PR intent: preserve multi-line [thinking] output in text mode instead of flattening agent_thought_chunk content into one line.
Status: the solution is right-shaped, targeted validation reproduced the regression and verified the fix, Codex review is clear, CI is green, and the final conflict check is clean.

Intent

Make the text-mode CLI output preserve multi-line agent thinking exactly enough to stay readable, instead of flattening it into one line.

Why

agent_thought_chunk text is currently passed through whitespace-collapsing formatting, so line breaks and blank lines turn into spaces and structured reasoning becomes hard to read. This PR updates the implicated flushThoughtBuffer() path in src/output.ts to preserve normalized newlines and adds a regression test in test/output.test.ts.

Codex review

No GitHub Codex review findings were collected for the current head. The established local Codex review reported no discrete correctness issue that would warrant a follow-up fix.

Targeted validation reproduced and confirmed the fix:

• Passed on the PR head.
• Failed after local-only ablation of the non-test code change.
• Passed again after restoring the PR branch state.

CI/CD

The initial CI workflow run was approval-blocked with action_required. Workflow approval was attempted and succeeded. The rerun of workflow 23142640688 completed successfully with all relevant jobs green. Initial and final conflict checks against origin/main were both clean.

Recommendation

This PR is ready for human landing decision. If the change aligns with maintainer expectations for text-mode thought rendering, it can proceed through normal human review and landing.