bug(vllm-performance): default values for reasoning_parser and tool_call_parser hinder experiment parametrization

[mgazz]

Issue Description

The test-agentic-tool-calling experiment sets default value for the following properties

• reasoning_parser
• tool_call_parser

Problem:

• Default values stop us from parametrising the properties as part of the experiment definition in a discovery space.
• We risk injecting unexpected parameters when properties are not defined in the space

Expected behaviour

• reasoning_parser and tool_call_parser should be omitted if not specified directly