fix logmean bound

[sbfnk]

d0bea23 introduced a bug where the logmean of delays is lower-bounded by 0 whereas it shouldn't be (for the gamma distribution this is done in the delays_lp function).

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 1023267 is merged into main:

•   :ballot_box_with_check:default: 1.15m -> 1.08m [-16.44%, +5.5%]
•   :ballot_box_with_check:no_delays: 1.07m -> 1.02m [-15.61%, +5.26%]
•   :ballot_box_with_check:random_walk: 21.1s -> 20.1s [-34.67%, +24.78%]
•   :ballot_box_with_check:stationary: 38.4s -> 39.9s [-9.53%, +17.36%]
•   :ballot_box_with_check:uncertain: 1.4m -> 1.57m [-8.73%, +32.97%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[sbfnk]

No NEWS item needed IMO as no change to previous minor version.

[seabbs]

LGTM. Good spot.