memory leak on repeated observe of an array of times

[kenta2]

Simplified testcase:

import numpy
from skyfield.api import load

eph = load('de440s.bsp')
timescale=load.timescale()

earth=eph['earth']
moon=eph['moon']
times=timescale.tt_jd(numpy.linspace(2400000,2500000,100000))
for i in range(1,500):
    obs=earth.at(times).observe(moon).apparent()
    print(i)

On my machine, this takes 10 minutes, and memory use keeps climbing, ending at 4 GB.

Skyfield 1.54 installed via pip install. Python 3.13.12 (Debian Forky Linux on amd64).