follow-up commit to https://gerrit.libreoffice.org/#/c/9226/ .

text_cat -s was documented as "Not very efficient yet, because
language models are re-loaded after each line." So if we want
to use text_cat -s for thousands of lines, better
read the language models only once and store them.

When tested on svl/source, the speed-up was a futher factor 1.5
(reduced from 6 s to 4 s).

Change-Id: I654a250b0e369e01c5eac5970b64df1390f0ef35
Reviewed-on: https://gerrit.libreoffice.org/9227Reviewed-by: Michael Meeks <michael.meeks@collabora.com>
Tested-by: Michael Meeks <michael.meeks@collabora.com>