Fix system-message tools parser when tool call is non-terminal

[MumuTW]

Summary

• stop treating a completed system-message tool call as terminal for the entire stream
• reset parser state after each completed tool call so later chunks/messages continue to be processed
• preserve assistant content that appears after a parsed tool call
• add regression coverage for post-tool content and mid-message tool-call placement

Testing

@continuedev/core@1.1.0 vitest /home/opc/.paperclip/instances/default/workspaces/7948d02f-b91e-4189-b9eb-32bf0b5923d2/continue/core
vitest run core/tools/systemMessageTools/toolCodeblocks/interceptSystemToolCalls.vitest.ts

 ELIFECYCLE  Command failed.
 WARN  Local package.json exists, but node_modules missing, did you mean to install? (fails in this workspace because is missing and is unavailable)

---

Continue Tasks: 🔄 7 running — View all

---

Summary by cubic

Fix system-message tool-call parsing so completed calls don’t end the stream and text before/after a call is preserved, including mid-message calls. Also dedupe dynamic tool definitions, emit trailing newlines as text, and map "reasoning-delta" to reasoning_content.

• Bug Fixes
  • System tools: make tool calls non-terminal; reset state after each call; preserve surrounding text; emit the trailing newline after a closed tool-call codeblock; avoid duplicate predefined tools in the dynamic section; add and correct tests for post-tool and mid-message cases.
  • OpenAI adapter: map "reasoning-delta" to delta.reasoning_content; simplify a test-only TS assertion.

^{Written for commit 97eeedc. Summary will update on new commits.}

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[cubic-dev-ai]

No issues found across 4 files

[MumuTW]

I have read the CLA Document and I hereby sign the CLA

[RomneyDa]

@MumuTW I think looks good, just a nitpick on unrelated changes

[RomneyDa]

This change looks unrelated to PR description, could you remove this and the test file?

[MumuTW]

Good catch — removed in ee012d3. Will open a separate PR for the reasoning-delta fix.

[MumuTW]

@RomneyDa Just checking in — the unrelated changes have been removed per your feedback (ee012d3). Let me know if anything else needs adjustment!

[MumuTW]

Fixed the failing core-checks CI — the two post-tool-call content tests had incorrect iteration counts:

• preserves content after a tool call: The loop consumed 6 iterations but only 4 are tool-call deltas. The trailing "\n" from \``\n` and the preserved text were being consumed inside the loop. Fixed by reducing the loop to 4 and adding explicit assertions for the newline and preserved content chunks.
• parses a tool call that appears mid-message and preserves trailing content: Added the missing assertion for the "\n" chunk between the closing code fence and "After tool".

All 46 system-message-tools tests pass locally. Commit: 97eeedc.