Changes in version 0.4.0                        

  - Adds maxent_model() for directed maximum-entropy ensembles
    constrained by connectance, link count, or degree sequences.
  - Supports directed unipartite food webs, optional self-links, locally
    isolated random streams, and optional retention of ensemble members.
  - Extends ecological_network objects with explicit directed and
    self_links fields.
  - Adds validation, unit tests, and a reproducible maximum-entropy
    vignette.

                        Changes in version 0.3.0                        

  - Renames the package and framework from rangerast to MetaWeave.
  - Positions the package around weaving regional ecological information
    into local spatial networks.
  - Adds probability_matrix_model() for direct use of named species-pair
    probability matrices, including ensemble predictions from external
    models.
  - Adds a self-contained methods and simulation vignette covering
    inference, spatial reconstruction, sampling error, and standardized
    metrics.

                        Changes in version 0.2.0                        

  - Generalizes bipartite inference from hard-coded palm and mammal
    groups to arbitrary named ecological groups.
  - Adds block_model() and local_block_network() as the generic
    block-model API.
  - Adds reconstruct_networks() and run_spatial_inference() as generic
    end-to-end workflows.
  - Ecological networks now record their row and column group
    identities.
  - Network summaries now expose generic row_richness and
    column_richness metrics.
  - Keeps sbm_model(), local_probability_network(), and the original
    downscaling wrappers for backward compatibility.
  - Fixes network-index construction when coordinate values carry names.

                        Changes in version 0.1.0                        

  - Introduces model-agnostic objects for distributions, assemblages,
    inference models, ecological networks, network collections, and
    spatial results.
  - Adds spatial assembly and range-rasterization utilities.
  - Adds the stochastic block model inference backend.
  - Adds simulation, network summaries, spatial mapping, and
    compatibility wrappers for the original downscaling workflow.