Interesting parts of the bugdoc:

- it has a numbering definition with two levels
- first level's type is none
- second level has a numbering string: "\x01."

Usually these placeholder bytes in the numbering string start from 0x00,
but there it starts at 0x01, which means the layout has to replace it
with the numbering from the second level.

Mapping from the spec to the code:

- nLevelB is an index into rgbxchNums
- aOfsNumsXCH is rgbxchNums
- sNumString is xst

So when the rNotReallyThere added in commit
251ba90d (INTEGRATION: CWS
soberfilterteam06 (1.44.26); FILE MERGED, 2003-05-19) wants to clear out
indexes from aOfsNumsXCH, it talks about numbering levels. The old code
assumed that nLevelB is the same as nPosValue, which is true in many
cases (when the levels are like 1, 1.1, 1.1.1), but not in this
particular case, where nLevelB is 0, but nPosValue is 1.

Change-Id: I590d9b2725a3330c26a04a526ce22d95970a974f
Reviewed-on: https://gerrit.libreoffice.org/32220Reviewed-by: Miklos Vajna <vmiklos@collabora.co.uk>
Tested-by: Jenkins <ci@libreoffice.org>