use-after-free in REGRESSION

[Ben Pfaff]

Valgrind reports the following for test 493 "REGRESSION" at
regression.at:3:

==28577== Invalid read of size 4
==28577==    at 0x494B263: regression_trns_resid_proc (regression.c:151)
==28577==    by 0x4A16C91: trns_chain_execute (transformations.c:197)
==28577==    by 0x49F6FC7: proc_casereader_read (dataset.c:530)
==28577==    by 0x49EE378: casereader_read (casereader.c:71)
==28577==    by 0x49ED399: buffer_case (casereader-shim.c:88)
==28577==    by 0x49ED594: casereader_shim_read (casereader-shim.c:111)
==28577==    by 0x49EE1E9: random_reader_peek (casereader.c:585)
==28577==    by 0x49EE7F3: casereader_peek (casereader.c:176)
==28577==    by 0x49EE887: casereader_is_empty (casereader.c:199)
==28577==    by 0x49EC1D5: casegrouper_get_next_group (casegrouper.c:113)
==28577==    by 0x4979891: cmd_list (list.q:128)
==28577==    by 0x48FEEB4: cmd_parse_in_state (command.c:222)
==28577==    by 0x48FF347: cmd_parse (command.c:152)
==28577==    by 0x804A5B6: main (main.c:139)
==28577==  Address 0x9701f48 is 16 bytes inside a block of size 140 free'd
==28577==    at 0x48DBB6A: free (vg_replace_malloc.c:366)
==28577==    by 0x4993DB3: linreg_free (linreg.c:125)
==28577==    by 0x494AF3D: cmd_regression (regression.c:408)
==28577==    by 0x48FEEB4: cmd_parse_in_state (command.c:222)
==28577==    by 0x48FF347: cmd_parse (command.c:152)
==28577==    by 0x804A5B6: main (main.c:139)

Maybe this is something introduced by the recent memory leak
fixes (I didn't check)?