fix order in public transport analysis

[nlehuby]

put back the offset (to make sure the projection on the the ways is ok even if the same way is used multiple times in the relation)

Fix #1262

[frodrigo]

Cannot be back as is. Was removed because of massive performances issues.

Eg try on http://download.openstreetmap.fr/extracts/south-america/brazil/southeast/rio-de-janeiro.osm.pbf

[nlehuby]

time docker-compose --project-name brazil_rio_de_janeiro run --rm backend ./osmose_run.py --country=brazil_rio_de_janeiro --skip-download --no-clean --analyser osmosis_relation_public_transport gives me

real	7m54,332s
user	0m0,710s
sys	0m0,142s

which seems reasonable :thinking:

[frodrigo]

It is probably linked to postgis/postgres or under lib version.

Look at this run 260/261 was killed after hours of timeout: https://buildbot.osmose.openstreetmap.fr/#/builders/371

I was able to reproduce the issue on osm110 debian with postgres 11.10, postgis 2.5.