sdbuildR 2.0.0

Breaking changes

• stockflow() is now the constructor for stock-and-flow models. Code that
  used xmile() or the development-only sdbuildR() constructor should call
  stockflow() instead.

• update() now replaces the old build() workflow for general model edits.
  For clearer model-building code, use stock(), flow(), constant(),
  aux()/auxiliary(), lookup(), and custom_func().

• stockflow is now the primary model class. Code that checked for
  sdbuildR_xmile or sdbuildR should check for stockflow instead.

• simulate() now returns simulate_stockflow objects, ensemble() returns
  ensemble_stockflow objects, verify() returns verify_stockflow objects,
  summary() returns summary_stockflow objects, and compare_models()
  returns compare_stockflow objects.

• sim_settings() now replaces sim_specs(), and meta() now replaces
  header().

• summary() now runs model diagnostics. Code that used debugger() should
  call summary() instead.

• dependencies() now replaces find_dependencies().

• export_model(format = "sdbuildR") now replaces get_build_code().

• sim_methods() now replaces solvers().

• custom_func() now replaces macro() for user-defined model functions.

• import_insightmaker() now replaces insightmaker_to_sfm(), and
  url_to_insightmaker() now replaces url_to_IM().

• ensemble() now uses conditions instead of range. To retain individual
  simulations, use save_sims = TRUE instead of return_sims = TRUE.

• plot() methods now use show_constants instead of add_constants. For
  ensemble plots, use which, sim, condition, and label_subplots instead
  of type, i, j, and j_labels.

• use_julia(nthreads = ) now controls Julia threading. Code that used
  use_threads() should set the thread count with use_julia() instead.

• Units are no longer supported. This means that unit-specific helpers from 1.x,
  including u(), convert_u(), drop_u(),
  model_units(), unit_prefixes(), get_units(), get_regex_units(), and
  get_regex_time_units(), are no longer available.

New features

• stockflow() creates an empty model or loads a built-in template using a
  case-insensitive template name.

• stock(), flow(), constant(), aux()/auxiliary(), lookup(), and
  custom_func() provide typed model-building helpers around update().

• update() now supports bare variable names and R expressions in name,
  type, eqn, to, from, and source. Strings and programmatic injection
  with !! remain supported.

• change_name(), change_type(), and discard() provide focused helpers for
  renaming, changing, and removing model variables.

• sim_settings() gains flexible output controls with save_at, save_n,
  vars, and save_sims.

• simulate() accepts temporary simulation-setting overrides through ...,
  using the same arguments as sim_settings().

• ensemble() can run ensembles in R and Julia, vary stocks and constants with
  conditions, and retain individual simulations with save_sims = TRUE.

• unit_test(), unit_tests(), and verify() add model-level unit tests for
  expected simulation behavior, including tests under alternative conditions.

• compare_models() compares model structure, equations, simulation settings,
  and nonlinearity scores across two stockflow models.

• import_desolve() converts deSolve-style ODE models into stockflow models.

• import_insightmaker() imports Insight Maker models from public URLs,
  .InsightMaker files, and ModelJSON .json files.

• export_model() exports models as sdbuildR reconstruction code, standalone
  deSolve scripts, or Psychomodels JSON records.

Improvements and fixes

• as.data.frame() methods now support the stockflow, simulate_stockflow,
  ensemble_stockflow, and verify_stockflow classes.

• dependencies() can now filter dependencies by variable name and type.

• export_plot() now supports the current plot classes and additional graph
  export options.

• simulate() and Julia code generation now use a faster AST-based R-to-Julia
  equation translator. Namespaced calls such as base::sum() translate
  correctly, and unsupported constructs fall back to the previous translator.

• simulate(), compilation, and summary() now use a content-hashed assembly
  cache. Repeated calls on unchanged models can skip reassembly, and editing one
  equation only retranslates that equation.

• simulate() now keeps Julia stock derivative positions aligned with the state
  vector when stocks are added, removed, renamed, or created by changing a
  variable's type.

• summary() and code generation now use an internal structural validator to
  catch inconsistent model layouts before simulation.

• simulate() now removes temporary CSV files created by single Julia
  simulations.

• textutils is now a required dependency.