This commit adds support for specifying context size when running benchmarks, which is passed to the Ollama client as the `num_ctx` option. The changes include:
- Updated the `run` method in the base benchmark class to accept an optional `context_size` parameter
- Modified the Ollama client call to include context size in the options when provided
- Updated the `run_benchmarks` function to accept and pass through the context size
- Added example usage to the help output showing how to use the new context size parameter
- Fixed prompt formatting in the summarization benchmark to use `text` instead of `task`
The changes enable running benchmarks with custom context sizes, which is useful for testing models with different context window limitations.
