Update persistent state document in the index the document belongs to

[przemekwitek]

Currently, the persistent state document is always indexed into the index pointed by an alias ".ml-state-write".
We plan this alias->index mapping to change over time (as we introduce rollover).
Therefore, we need to make sure we don't end up with two copies of the state document.
This PR achieves that by:

• searching for the doc
• either indexing it into the current write index or updating it in its current index

In order to know the document's id, bytes from C++ stream must be parsed (but only until the first new line character).

Relates #29938

[elasticmachine]

Pinging @elastic/ml-core (:ml)

[droberts195]

This looks like a dangerous change. Previously if the C++ wrote nothing but whitespace it was silently ignored. Now it will cause the state processor to throw an exception and never process any subsequent state.

I think the behaviour should be changed back to what it was for the pure whitespace case. It's fine to throw an exception if a bulk metadata JSON object is invalid. But if it's not present at all then I think we should maintain the previous behaviour of treating it as a no-op. Otherwise a very careful audit of the C++ code will be required to find what situations it might write blank state.

[przemekwitek]

Good finding, done.
Please note that I'm now skipping leading blank lines just in case the actual request body is present after those blank lines.

[droberts195]

LGTM but please can you add some comments to make it clearer to future maintainers why we are doing this

[droberts195]

LGTM