Fix the OOM problem introduced by low ke_cutoff

[henryw7]

When ke_cutoff is low, all shells with high ke will be placed into the same bucket (task), and that particular bucket requires a large amount of GPU memory for the Fock matrix in diff_image space.

[sunqm]

Please add a test for the limited memory case

with lib.temporary_env(multgrid_v2, get_avail_mem=(lambda **kw: 1000000)):
    ...