`Weibull` returns NaN from `pdf`/`cdf`/`sf`/`ln_pdf` for finite `shape` and `scale`

[DRMacIver]

use statrs::distribution::{Continuous, ContinuousCDF, Weibull};

fn main() {
    for (shape, scale) in [(2000.0, 0.5), (1e6, 0.99), (1500.0, 0.9)] {
        let w = Weibull::new(shape, scale).unwrap();
        println!("Weibull({shape}, {scale}): pdf = {}, cdf = {}, sf = {}, ln_pdf = {}",
            w.pdf(0.4), w.cdf(0.4), w.sf(0.4), w.ln_pdf(0.4));
    }
}

Output:

Weibull(2000, 0.5): pdf = NaN, cdf = NaN, sf = NaN, ln_pdf = NaN
Weibull(1000000, 0.99): pdf = NaN, cdf = NaN, sf = NaN, ln_pdf = NaN
Weibull(1500, 0.9): pdf = 0, cdf = 0, sf = 1, ln_pdf = -1208.1658132055286

new accepts all three. A density is never NaN, and Weibull(1500, 0.9) shows the neighbouring case coming back finite.

Possibly a duplicate of #310, which describes the same NaN in the same expression. Filing separately because #310 is scoped to infinite parameters throughout — its repro passes f64::INFINITY and the discussion turns on rejecting those at construction, which would leave this case, where every parameter and the argument are finite. Happy for it to be folded into #310.

Tested on statrs 0.19.1 and current main (50fcf4d).

(This was another bug found using hegel.)