Necessary fixes for _bulk endpoint

[mieciu]

For our V2 ingest processor, we need better implementation of _bulk API, which can handle more situations. Some shortcuts we took earlier, are no longer a viable option.

Changes:

• We used to take bulk payload lines and just process them in pairs (action/metadata line followed by document) which doesn't have to be the case. There can always be a delete call which isn't followed by any document. So the whole BulkForEach had to be rewritten.
• Extending NDJSON type validation, so that single-line JSON is also a valid NDJSON 😉
• Adding some tests

When loading sample datasets and having operational Kibana, every single request ends with code 200 🎊

[mieciu]

IT run https://github.com/QuesmaOrg/quesma/actions/runs/12871349933

[mieciu]

Had to retry previous one and it succeeded, looks like a flakiness, weird...
https://github.com/QuesmaOrg/quesma/actions/runs/12871859960