Feature: Add sampling parameters to centrality metrics for faster computation

[Robinlovelace]

This feature request proposes adding a sampling mechanism to the centrality calculation functions (local_node_centrality_shortest, local_node_centrality_simplest, and local_segment_centrality).

By introducing a sample_probability parameter, users can opt to run the expensive centrality computations on a random subset of nodes. Additionally, a weighted_sample parameter allows this sampling to be biased by node weights (e.g., population or trip generation potential).

This functionality allows for significantly faster approximate centrality calculations, which is particularly useful for large networks or iterative workflows where exact precision for every node is not initially required.

I have implemented this in the Rust backend and exposed it via the Python API.

[songololo]

Released in the 5.x line as opt-in adaptive sampling:

cn.centrality_shortest(distances=[...], sample=True) (add epsilon= to set the error tolerance; default 0.05).

Sampling rates are set per distance threshold from each node's measured reach (Hoeffding bound), and results are corrected by inverse-probability weighting so estimates stay unbiased. A per-distance work test also falls back to exact computation when sampling wouldn't be cheaper. On the weighted_sample idea — node weights are already respected in the estimation.

Docs: https://cityseer.benchmarkurbanism.com/metrics/sampling
Guide: https://cityseer.benchmarkurbanism.com/guide/centrality#adaptive-sampling
Recipe: https://cityseer.benchmarkurbanism.com/examples/centrality/sampled-centrality

Closing as implemented.

[Robinlovelace]

Great job on the fix @songololo !