Inconsistent use of `_` to separate words in method names

[Anaphory]

Sometimes, method names contain underscores to separate words (to_string) and sometimes not (tolist). This catches me of regularly. Particulary annoying are the cases where the same prefix (eg. to) sometimes takes an underscore and sometimes not.

Is this going to change for future major versions?

Example: dir(Index)

• Without underscore between words, but the compound is well-established: 'argmax', 'argmin', 'argsort', 'nbytes', 'ndim', 'nlevels', 'nunique', 'putmask'
• Without underscore: 'astype', 'fillna', 'hasnans', 'isin', 'itemsize', 'searchsorted', 'sortlevel', 'tolist'
• With underscore: 'drop_duplicates', 'get_duplicates', 'get_indexer', 'get_indexer_for', 'get_indexer_non_unique', 'get_level_values', 'get_loc', 'get_slice_bound', 'get_value', 'get_values', 'groupby', 'has_duplicates', 'holds_integer', 'inferred_type', 'is_all_dates', 'is_boolean', 'is_categorical', 'is_floating', 'is_integer', 'is_lexsorted_for_tuple', 'is_mixed', 'is_monotonic', 'is_monotonic_decreasing', 'is_monotonic_increasing', 'is_numeric', 'is_object', 'is_type_compatible', 'is_unique', 'memory_usage', 'set_names', 'set_value', 'slice_indexer', 'slice_locs', 'sort_values', 'sym_diff', 'symmetric_difference', 'to_datetime', 'to_native_types', 'to_series', 'value_counts'

[jreback]

mostly a dupe of #8826

the only ones with any rationale are:

tolist -> to_list (for compat with rest of pandas to_* methods) hasnans -> hasna (inconsistent naming, e.g. fillna, dropna)

the other are long long established and in particular searchsorted, itemsize, astype are super common numpy references

sortlevel should actually be deprecated as sort_index does all of this.