Here are two proposed solutions:

1. Have a single JSON file with the collection of all titles. The file size may reach several hundred GBs.
2. Split data into multiple files that can fit memory of a regular computer. Each file is going to be presented as a paginated version having the total number of pages, total number of titles, current page and the number of titles on this page. Each file size is going to be about 10GB and the total number of files may reach 100.

#1 seems like not an option :)

For #2...what would that that paginated version look like?

We can follow the approach we have for diffs where it would be a set of files named with the wiki name and the incremental number

enwiki_1.json
enwiki_2.json
...

The content of files will be similar to diffs:

[
 {title1},
 {title2},
  ...
]

We want this to be in njson format.

Meaning without top level array, so that this can be easily parsed line by line.

{...}
{...}
{...}

If we are to keep file size under certain limit for single file, let's say 10GB, then this approach to naming files makes the most sense to me:

enwiki_1.json
enwiki_2.json
...