UPSTREAM PR #17924: Gigachat 3 tool parser and tests

[loci-dev]

Mirrored from ggml-org/llama.cpp#17924

This PR adds a tool parser for parsing the function calls made by the Gigachat 3 models.
Models can generate only one function call using format:

{text}<|message_sep|>\n\nfunction call<|role_sep|>\n{json}
Where <|message_sep|>\n\n and <|role_sep|>\n are special tokens.

[loci-review]

Explore the complete analysis inside the Version Insights

I've successfully generated the summary report for your project. The report shows a performance analysis comparing two versions of the llama.cpp repository (PR #520) from auroralabs-loci.

Key Highlights:

• Most Impacted Function: The begin function shows a dramatic 214% increase in response time and 289% increase in throughput
• Affected Binaries: Primarily llama-run and llama-cvector-generator
• Overall Pattern: All top 10 functions show increases in both response time and throughput, suggesting more frequent invocations in the new version
• Main Categories: STL container operations, memory management, smart pointers, and template utilities

The report includes detailed metrics for each of the top 10 functions and provides recommendations for further investigation, particularly around the begin function and memory allocation patterns.