Reset per compressed row memory context during tuple filtering

When a batch was skipped due to compression tuple filtering the
per compressed row memory context would not be reset leading to
memory usage increasing linearly with number of consecutive filtered
batches.

(cherry picked from commit 460994c)

.unreleased/pr_7270

@@ -0,0 +1 @@
+Fixes: #7270 Fix memory leak in compressed DML batch filtering

tsl/src/compression/compression.c

@@ -1457,6 +1457,7 @@ build_decompressor(Relation in_rel, Relation out_rel)
 void
 row_decompressor_reset(RowDecompressor *decompressor)
 {
+	MemoryContextReset(decompressor->per_compressed_row_ctx);
 	decompressor->unprocessed_tuples = 0;
 	decompressor->batches_decompressed = 0;
 	decompressor->tuples_decompressed = 0;
@@ -1791,8 +1792,6 @@ row_decompressor_decompress_row_to_table(RowDecompressor *decompressor)
 	}
 
 	MemoryContextSwitchTo(old_ctx);
-	MemoryContextReset(decompressor->per_compressed_row_ctx);
-
 	row_decompressor_reset(decompressor);
 
 	return n_batch_rows;
@@ -1812,8 +1811,6 @@ row_decompressor_decompress_row_to_tuplesort(RowDecompressor *decompressor,
 	}
 
 	MemoryContextSwitchTo(old_ctx);
-	MemoryContextReset(decompressor->per_compressed_row_ctx);
-
 	row_decompressor_reset(decompressor);
 }