perf(decompress): Batch column drops in decompress_columns()

Individual df.drop() per column caused O(N_cols × N_rows) pandas
reindex on each call. For 7 columns × 82M rows, this was 12s out
of 17s total (72% of runtime).

Fix: Collect all compressed columns, drop once after loop.
Expected: 17s → ~5s for 7-column decompression on 82M rows.

Profile-driven: cProfile showed df.drop → _slice_take_blocks_ax0
→ numpy.take consuming 12.1s across 7 calls.

Author: miranov25

UTILS/dfextensions/AliasDataFrame/AliasDataFrame.py

@@ -6692,6 +6692,10 @@ def decompress_columns(self, columns=None, inplace=False, keep_compressed=True,
         # Filter __meta__
         columns = [c for c in columns if c != "__meta__"]
 
+        # Phase 13.7: Collect columns to drop, then batch-drop once at end
+        # (Avoids O(N_cols × N_rows) pandas reindex per drop)
+        cols_to_drop = []
+
         for col in columns:
             if col not in self.compression_info:
                 raise ValueError(
@@ -6741,9 +6745,9 @@ def decompress_columns(self, columns=None, inplace=False, keep_compressed=True,
             if col in self.aliases:
                 del self._schema["columns"][col]
 
-            # Step 4: Handle compressed column
+            # Step 4: Collect compressed column for batch drop
             if not keep_compressed:
-                self.df.drop(columns=[compressed_col], inplace=True)
+                cols_to_drop.append(compressed_col)
 
             # Step 5: Update state
             if keep_schema:
@@ -6753,6 +6757,10 @@ def decompress_columns(self, columns=None, inplace=False, keep_compressed=True,
                 # Remove all compression metadata
                 del self.compression_info[col]
 
+        # Batch drop all compressed columns at once (single reindex)
+        if cols_to_drop:
+            self.df.drop(columns=cols_to_drop, inplace=True)
+
         return self
 
     def get_compression_info(self, column=None):