High memory consumption when getting address book cards

[juliusknorr]

Filing here as an issue to collect findings, I'm not entirely sure if that is something we really can solve properly e.g. by streaming the individual card data rather than keeping them all in memory.

There seems to be at least some additional overhead from #19901 but the major impact is the database result that is loaded.

https://blackfire.io/profiles/93136cfc-7b78-4919-b8ae-5695c0e905e2/graph?settings%5Bdimension%5D=pmu&settings%5Bdisplay%5D=landscape&settings%5BtabPane%5D=nodes&selected=OCA%5CDAV%5CCardDAV%5CCardDavBackend%3A%3AgetCards&callname=main()&constraintDoc=

Some useful queries:

select id, sum(char_length(carddata)) from oc_cards where addressbookid = 7;
select id, char_length(carddata) as size from oc_cards where addressbookid = 7 order by size;
select substring_index(carddata, '\n', 10) from oc_cards where id = 3296;