Batch preprocessing: weighting options and API refactor

[kgdunn]

Migrated from TODO.md (Batch section).

Items covered

• Refactor batch preprocessing to a scikit-learn-style .fit() / .apply() API. (batch/preprocessing.py)
• Handle the DataFrame input cases. (batch/preprocessing.py)
• Revisit thesis page 181: handle multiple points in the target. (batch/preprocessing.py)
• Document the function completely. (batch/preprocessing.py)
• Try sum-of-absolute-values weighting alongside the quadratic option. (batch/preprocessing.py)
• Allow leaving out the worst batches when computing weights. (batch/preprocessing.py)
• Make the weighting choice a configurable setting. (batch/preprocessing.py)

Note: code line numbers in the old TODO.md are stale; locate the code by symbol name.