Automatically use old and new style node creation.

Don't use ##args style in macros

It was times when there were no tuplesort_free() at all

…_getindextuple

This reverts commit d9963d9.

- fix write uninited values in logical tape
- fix double free

It was incorrectly set different RumSortSize vs RumScanSize

Copy item additional information from leaf data page. It is necessary when
additional information is used after the data page is unlocked.

In paricular when sorting of items should be done.

- Move code duplicates into separate functions
- Disable import of trace_sort from backend ads it is unconditionally
  defined in the module (in tuplesortXX.c)

- remove unused defines
- use tuplesort_begin_common instead of rum_tuplesort_begin_common
- simplify version-conditional defines
- rearrange functions order in rumsort.c

- Make rumsort use vanilla tuplesort functions. They are included in tuplesortXX.c which are copied without change from vanilla XX sources (src/backend/utils/sort/tuplesort.c)
- Add compatibility with PG15 logtape changes

-Add vanilla tuplesort.c and qsort_tuple.c files

Fix typo

information is not added.

between write and replay on replica. Now test will
wait for reply on target database not write to wal
tags: rum

8524bf0

tags: rum

…thout

positional information (e.g. addon_ops).

Per report: #93

tags: rum

due to adding padding bytes by Postgres data access alignment macros.
This was the source of rare but dangerous segfault on 32-bit FreeBSD
but no system was safe as static alignment is completely
system/compiler free choice.

This problem was hidden by the added completely unrelated variable
trace_sort way before the relevant part of the code. It just shifted
the alignment of all variables with bigger address values to
acceptable but haven't solved the problem at large.

( related to 338c4e1 )