SelectBoost.quantile 0.3.0

Methodology

Added hybrid support extraction as the default summary/support rule, combining path stability and fitted effect size.
Regenerated the shipped validation study with the improved selector and the strengthened high-dimensional safeguards.

Added a getting-started vignette for the current matrix, formula, tuning, and prediction workflows.
Updated the validation vignette to report F1 score alongside recall and false-discovery rate.
Updated the README to point to the new vignettes and current benchmark defaults.

Updated inst/scripts/run_quantile_benchmark.R so benchmark artifacts are written to a user-supplied directory or, by default, to a subdirectory of tempdir() rather than the package tree or getwd().
Reworked internal seed handling to use withr::with_seed() instead of touching .GlobalEnv, and added regression tests to confirm seeded helpers preserve the caller RNG stream.
Updated benchmark reproduction examples in the README and validation vignette to use temporary output directories.

Refactored the core SelectBoost-style quantile workflow into modular preprocessing, grouping, directional, tuning, and methods layers.
Added multi-tau support, formula interfaces, prediction helpers, support extraction, and broader lambda tuning workflows.
Added a validation framework with default benchmark scenarios, per-method support recovery summaries, failure tracking, and a reproducible benchmark runner.

Added a validation vignette based on shipped benchmark artifacts under inst/extdata/validation/.
Expanded README guidance around validation and reproducible study runs.