Supporting in-document IDs and joined doc imports/exports

[ashvardanian]

Benefits

Java, GoLang and many other bindings will receive "upsert" functionality with just a single char const * argument.
Similarly, streaming exports can emplace ID into the packed document, to simplify post-processing for user.
This form is compatible with Mongo DB and Elastic stack, which are behind in terms of Apache Arrow adoption.

Changes

1. If no docs_count is set:
   • if the format is JSON - we count the newlines.
   • we need to have at least the first length variable set.
2. Every input document is checked to contain an integer-castable top-level _id field.