feat(examples): Add Information Cascade model to benchmark memory latency (Relates to #200)

[ZhehaoZhao423]

What's the problem this feature will solve?
Following the community's ongoing discussions regarding exponential API latency and event-loop blocking (specifically raised by @crocmons in #200), there is a need for a reliable, reproducible benchmark to measure how memory modules behave under high-frequency, multi-agent interactions.
While #200 excellently identifies the execution-layer overhead, my testing shows that the underlying STLTMemory consolidation logic inherently exacerbates this bottleneck by forcing blocking LLM calls whenever the deque fills up.

Describe the solution you'd like
I propose adding a new Agent-Based Model example to the examples/ directory: Information Cascade (Financial Rumor Mill).
In this model, LLM-powered quantitative traders rely purely on market rumors and STLTMemory to make decisions. By deliberately constraining short_term_capacity=1, this model forces rapid, continuous memory consolidation, effectively stress-testing the memory architecture and bypassing default UI overhead.

Are you planning to open a PR for this?
Yes. I already have the model, a custom robust Solara UI (app.py), and headless benchmark scripts (run_benchmark.py) fully prepared and formatted with pre-commit locally. I will open a PR immediately following this issue.

Additional context
This example provides two massive values to the community:

1. Measurable Benchmarking for Bug: Critical Performance Issues - API Latency and Inefficient Parallel Execution #200: The headless script explicitly times the synchronous API blocking delay during step(), providing a strict baseline to measure the success of future concurrency or memory optimization PRs.
2. Behavioral Testing: It perfectly demonstrates "Echo Chamber" hallucinations and cognitive degradation when LLMs repeatedly summarize their own outputs over long simulation runs.