Validate shared log weight normalization

[fujikix1102]

Summary

• centralize max-scaled log-weight conversion with explicit finite and positive-sum checks
• use the contract in sampler chain starts, hints, GetDist export, and postprocessing weights
• preserve zero-weight samples while rejecting NaN, positive infinity, empty, and all-zero inputs

Validation

• PYTHONPATH=$PWD pytest -q cosmosis/test/test_utils.py cosmosis/test/test_samplers.py -k 'normalized_log_weights or maxlike' --disable-warnings --maxfail=1
• 19 passed, 33 deselected
• PYTHONPATH=$PWD pytest -q cosmosis/test/test_getdist.py --disable-warnings --maxfail=1
• 3 passed