rzig 0.2.3

CRAN resubmission

Removed redundant wording from the package title and description.
use_rzig() and document() now require an explicit package path, so neither function writes to the working directory by default.
Added executable, temporary-directory examples for both exported functions; they use \donttest{} because they require the external Zig toolchain.
document() now confines Zig’s local and global caches to an automatically cleaned R session temporary directory. On Unix, its compiler subprocess also receives a private temporary directory.

use_rzig() now scaffolds cleanup and cleanup.win, preventing generated Makevars and Zig build caches from entering source-package checks.
document() uses the same Zig discovery order as package configuration and rejects unsupported compiler versions before running the export scanner.
The onboarding guide now edits the generated example in place, documents the complete parameter and return surface, and explains warning and roxygen2 workflows.

Integer, logical, and character vectors can be returned directly, decorated with attributes, and placed in generated lists.
Ctx.rng() provides boundary-managed draws from R’s random-number stream, preserving set.seed() reproducibility on success and error paths.
Rmath.normalCdf() and Rmath.normalQuantile() expose the linked R runtime’s normal distribution functions without consuming random state.

User interrupts retain R’s interrupt condition class after Zig cleanup; ordinary error handlers and try() no longer swallow Ctrl-C.
The POSIX integration gate sends a real SIGINT, checks the condition class, and verifies that the R session remains usable.
The causal demonstration now returns a logical adjacency matrix and covers all three-variable orders plus a 12-variable search through depth five.

use_rzig() scaffolds a portable Zig-backed R package, including generated registration, build configuration, and a working native example.
document() derives R wrappers, exports, and documentation from plain Zig functions marked with /// @export.
Generated packages build on Linux, macOS, and Windows with Zig 0.16.0 and R’s platform toolchain.

Typed conversions cover numeric, integer, logical, optional, string, vector, list, matrix, attributed, and explicitly mutable values.
Mutable numeric inputs follow R copy-on-modify semantics by duplicating before Zig receives writable storage.
Long computations can check R interrupts, and pure Zig indexed loops can use capability-restricted worker threads.

Native errors and ReleaseSafe panics become R conditions after Zig cleanup has completed.
The boundary protects R allocations, keeps borrowed inputs read-only, and confines R API access to the calling thread.
Cross-platform checks include hostile-input tests, garbage-collection stress, valgrind, native interface analysis, and reproducible boundary benchmarks.

Added a worked package tutorial and a vignette showing how to port an Rcpp hot loop while preserving copy-on-modify, error, and interrupt behavior.