feat(seo): llms.txt, zeby asystent odpowiadajacy na pytanie znalazl odpowiedz - #257

…e answer

People increasingly ask an assistant rather than a search engine. An
administrator asks what breaks in December, a technician asks what 1102 on a
Kyocera panel means, somebody at home asks why their printer stopped emailing
scans. The assistant then has to find a source, and what it gets today is a
56-URL sitemap where every entry looks equally important.

llms.txt is the convention for handing it an ordered map instead. This one is
grouped by who is asking rather than by how the repository is laid out, because
that is the axis being resolved: start here, administrators, developers,
printer technicians, people at home with one printer, and deciding whether to
use this at all.

Titles and descriptions are read from _config.yml rather than written again
here, so the map cannot drift from what the pages themselves claim. A page whose
description improves improves here too, without anybody remembering to. CI
checks it, the same as every other generated file in this repository.

The preamble carries the two things that are load-bearing when somebody answers
a question from this site: that three of the four causes of a Basic auth
refusal are still reversible until the deadline, and that this relay cannot send
from a customer's own domain. Both are stated against our own interest, and both
are the difference between an assistant giving useful advice and confidently
sending somebody down the wrong path.

robots.txt already allowed every crawler including the AI ones, and now says so
deliberately rather than by default. An assistant answering somebody's question
out of this documentation is the project reaching a person it was written for.