fix(web-search): always stream upstream requests while buffering loop iterations

[orangeboyChen]

Problem

With local web search enabled, requests fail immediately:

{
  "msg": "Non-stream chat request is currently not supported",
  "code": 11101
}

Cause

The upstream /v2/chat/completions endpoint only accepts stream: true, and buildUpstreamBody sets exactly that. The web search loop, however, needs a complete tool call before it can run a query, so it overrode the request with stream: false — which upstream rejects with code 11101.

Fix

Follow the upstream contract instead: the request always uses stream: true, and the loop buffers the resulting SSE into a single complete message (reusing the existing aggregateUpstreamStream()) before inspecting it for tool calls. Non-streaming callers still receive one JSON payload, so behaviour is unchanged.

The stream parameter on fetchChatCompletion is now documented as controlling only the shape handed back to the caller (pass through SSE vs. buffer into JSON) — it no longer affects what is sent upstream.

Verification

• New regression test asserts every request sent upstream has stream: true. Confirmed it fails without the fix.
• 384 server tests pass
• lint / format:check / typecheck / build all pass
• test:ci coverage: 94.31% statements, 87.93% branches, 94.88% lines

Note

This is a follow-up fix to #127, which has already been squash-merged into main, so it needs its own PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.09%. Comparing base (012b3ea) to head (b3cdabb).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #130      +/-   ##
==========================================
+ Coverage   95.01%   95.09%   +0.08%     
==========================================
  Files          26       26              
  Lines        4950     4950              
  Branches     1432     1432              
==========================================
+ Hits         4703     4707       +4     
+ Misses        247      243       -4     

Flag	Coverage Δ
unittests	95.09% <ø> (+0.08%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b3cdabbc17

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve in-band SSE errors while buffering

When CodeBuddy reports a streaming failure as an HTTP 200 SSE frame such as data: {"error": ...}, this newly forced streaming path passes it to aggregateUpstreamStream, which ignores the error field and manufactures a successful 200 completion with empty content. The web-search loop therefore returns an apparent empty answer instead of propagating the upstream failure; detect and preserve SSE error payloads during aggregation.

Useful? React with 👍 / 👎.