Refactor Hermes3 tool parser streaming mode

[mzegla]

With that change hermes3 parser is capable of handling more complex chunks in streaming mode.
Additionally some parts of the logic have been wrapped in separate functions and processing phases in parseChunk are described in details in the comments.

[Copilot]

Pull Request Overview

This PR refactors the Hermes3 tool parser to improve its streaming mode capabilities by handling more complex chunks and reorganizing the processing logic. The changes add better support for edge cases and provide clearer separation of processing phases.

• Refactored streaming parser with improved chunk handling and state management
• Enhanced test coverage with additional tool call scenarios and edge cases
• Added proper error handling with stream cache clearing on exceptions

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
third_party/llm_engine/llm_engine.bzl	Updated commit date in comment from 2025-09-13 to 2025-09-23
src/test/llm/output_parsers/hermes3_output_parser_test.cpp	Extended test cases with new tool call scenarios and updated comment
src/llm/io_processing/output_parser.cpp	Added try-catch blocks to clear stream cache on parser exceptions
src/llm/io_processing/hermes3/tool_parser.hpp	Added new private methods for chunk processing and state management
src/llm/io_processing/hermes3/tool_parser.cpp	Complete refactor of parseChunk method with improved logic and detailed comments

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

The abbreviation 'i.e.' should be followed by a comma. It should be 'i.e., <tool_call>(text)*</tool_call>'.

Suggested change

	// We assume single chunk will not contain whole tool call i.e. <tool_call>(text)*</tool_call>
	// We assume single chunk will not contain whole tool call, i.e., <tool_call>(text)*</tool_call>