heavytails 0.2.0

New Functions

mle_pareto() — Parametric MLE for the Pareto tail index using the closed-form estimator from Theorem 8.1 of Nair et al. Supports an optional finite-sample bias correction (§8.3). Returns a heavytails_mle object with a print() method.
wls_pareto() — Weighted least-squares estimator for the Pareto tail index via log-log rank regression (Theorem 8.5 of Nair et al.). Optionally draws the rank plot with both WLS and OLS fitted lines. Also returns the unweighted OLS estimate for comparison.
ks_xmin() — Standalone KS-minimization estimator for the power-law lower bound x̂_m (Step 1 of the Clauset et al. pipeline, §8.5). Extracts the core loop from plfit() so x̂_m can be estimated without running the full power-law fit.

ks_gof() — Bootstrap Kolmogorov-Smirnov goodness-of-fit test for the Pareto distribution (Step 2 of the Clauset et al. pipeline, §8.5). The p-value is the fraction of parametric bootstrap KS statistics exceeding the observed statistic.
lr_test_pareto() — Vuong likelihood-ratio test comparing the Pareto fit against alternative distributions: "exponential", "lognormal", and "weibull" (Step 3 of the Clauset et al. pipeline; Clauset et al. 2009, §3.3). Returns a data.frame with the LR statistic, two-sided p-value, and preferred model for each alternative.

hill_plot() — Hill plot: α̂ vs. k over a range of tail sizes. Accepts an optional alpha_true argument to overlay the true value as a reference line (useful in simulation studies). Returns the plotted data.frame invisibly.
moments_plot() — Moments estimator plot: ξ̂ vs. k. Returns the plotted data.frame invisibly.
pickands_plot() — Pickands estimator plot: ξ̂ vs. k. Returns the plotted data.frame invisibly.
rank_plot() — Log-rank vs. log-x plot for assessing Pareto linearity. Returns the plotted data.frame invisibly.
qq_pareto() — Pareto Q-Q plot. Accepts an optional alpha argument; if omitted, mle_pareto() is called internally on the data. Returns the theoretical and empirical quantiles invisibly.

rpareto() — Generate Pareto(xm, α) random variates. Promoted from internal helper to exported function.
pareto_cdf() — Pareto CDF. Promoted from internal helper to exported function.
dpareto() — Pareto density function. New function.

Hill estimator indexing corrected. The denominator in the Hill formula was X_(k) instead of X_(k+1), causing the k-th term in the average to always be zero and biasing the estimate upward. The formula now correctly divides by X_(k+1), consistent with Eq. 9.12 of Nair et al. and the internal db_estimators() helper used by doublebootstrap().
Dead code removed from doublebootstrap(). The if (n2 < 5) check was unreachable because the preceding if (n2 <= 5) already covered that case. The if (n1 != -1 && n1 >= n) check was also unreachable because n1 had already been overwritten from its -1 default earlier in the function body. Both branches have been removed.

gpd_lg_likelihood() is no longer exported. It is an internal optimization helper for pot_estimator() and was never intended for direct use. Any code calling gpd_lg_likelihood() directly should switch to heavytails:::gpd_lg_likelihood() or, preferably, use pot_estimator() instead.

Added README.md with installation instructions, a quick-start example, a function reference table, and a full Clauset et al. pipeline walkthrough.
All new functions have full roxygen2 documentation with LaTeX math, parameter descriptions, return value specifications, examples, and references to Nair et al. (2022) and Clauset et al. (2009).
Added NEWS.md (this file).

Initial CRAN release.
Implemented tail index estimators: hill_estimator(), moments_estimator(), pickands_estimator(), pot_estimator(), plfit(), doublebootstrap().