Fast Tasks Dominated by OS Noise

`retrieval_core` (~0.002s) and `graph_analytics` (~0.007s) are so short that a single OS scheduler preemption (typically 1–10ms) can produce a 37%+ apparent delta. Their benchmark sizes are too small to produce stable timings

Refer https://github.com/iiitl/chuck/issues/8#issuecomment-4229800275 for more details

Start by reading the linked comment on issue 8 for the proposed approach, then locate the benchmark definitions for retrieval_core and graph_analytics. Check how their current sizes and timings are measured; done means the two fast tasks produce stable, representative benchmark results with reduced OS-noise effects.

Written by the indexing model from the issue text.