Fix bugs in shapeN + CMSHistFunc/Sum

[ajgilbert]

• Fixes major bug when using --X-rtd FAST_VERTICAL_MORPH in conjunction with CMSHistSum where the log value of the templates were applied in every evaluate, not just the first time
• Fix minor bug in CMSHistFunc and CMSHistSum which now explicitly fixes the normalization after shapeN interpolation to the nominal

[nsmith-]

In principle we could now require the test https://github.com/cms-analysis/HiggsAnalysis-CombinedLimit/actions/runs/10043200902/job/27755411724#step:6:475 to succeed:

Out of 10 toys, {'kl': 0, 'kv': 0, 'k2v': 0} are not matching (tolerance: 1e-4) between CMSHistFunc and CMSHistSum