BUG: Fix inconsistency when read_csv reads MultiIndex with empty values (#59560)

[allamlobna]

• closes BUG: inconsistency when read_csv reads MultiIndex with empty values #59560
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

Summary

This PR fixes an inconsistency where read_csv replaced empty
MultiIndex column level values with automatically generated
"Unnamed: x_level_y" placeholders.

Empty values are now preserved as empty strings (""),
matching MultiIndex index behavior and ensuring consistent
roundtrip results between to_csv and read_csv.

Changes

• Added _clean_column_levels() helper to normalize MultiIndex
  column labels in BaseParser.
• Updated _extract_multi_indexer_columns() to use it.
• Adjusted test_multi_index_unnamed() expectations.
• Added regression test for GH#59560 in test_header.py.
• Added whatsnew entry under Bug Fixes → IO.

Impact

• Aligns column + index behavior for MultiIndex CSVs.
• No change for single-level headers.
• Both C and Python parsers tested successfully.

[allamlobna]

Hi, this PR currently fixes GH#59560, which addresses inconsistent handling of empty MultiIndex level values in read_csv.

Summary of changes:

• Added _clean_column_levels() to normalize empty or automatically generated "Unnamed: x_level_y" placeholders when reading MultiIndex columns.

• Hooked this into _extract_multi_indexer_columns() in the CSV parser so empty header cells are preserved as "" instead of "Unnamed:".

Observed side effect:
Because this normalization happens in shared column-cleaning logic, it’s also affecting other I/O readers which now return empty strings instead of "Unnamed:" placeholders.
This causes several tests in those modules to fail since they expect the old "Unnamed:" behavior.

Question for maintainers:
Would you prefer that I:

• Limit the scope to CSV in this PR (restore previous behavior for Excel/HTML to keep this focused on GH#59560),

or

• Extend the change across all I/O readers and update the corresponding parser tests for consistency in this same PR?

I’m happy to take either direction, just wanted to check what’s preferable from a review and scope standpoint.