Fixing content parsing for onyx

[przepeck]

🛠 Summary

Removing to=user<|message|> from content with new version of model

🧪 Checklist

• Unit tests added.
• The documentation updated.
• Change follows security best practices.
  ``

[Copilot]

Pull request overview

This PR updates the Onyx LLM streaming output parsing behavior to strip model framing tokens (e.g., to=user / <|message|>) from content deltas so clients don’t see those protocol markers in streamed responses.

Changes:

• Updated the Onyx streaming content-only unit test to include <|start|> and assistant as separate initial chunks.
• Adjusted Onyx tool parser “special tags to erase” to also erase <|message|>, and changed the content-start indicator tokenization.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/test/llm/output_parsers/onyx_output_parser_test.cpp	Adjusts streaming test input chunks to match updated model framing emission.
src/llm/io_processing/onyx/onyx_tool_parser.hpp	Updates constants and the tag-erasure list used during streaming content parsing.
src/llm/io_processing/onyx/onyx_tool_parser.cpp	Updates the concrete tag strings for content framing removal.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/test/llm/output_parsers/onyx_output_parser_test.cpp:677

• These added chunks do not cover the regression: the previous combined to=user<|message|> tag already buffers and removes these adjacent chunks, so the test passes without the production change. Put the completed assistant prefix and to=user in the same chunk; under the old code, finding ASSISTANT_PREFIX caused the cache to flush the still-incomplete to=user<|message|> envelope and leak to=user, while the new separate tags prevent that.

        {"<|start|>", ov::genai::GenerationFinishReason::NONE, std::nullopt},
        {"assistant ", ov::genai::GenerationFinishReason::NONE, std::nullopt},