Target optimal num docs by split when ingesting data

[fmassot]

We can ingest data from a file with the CLI. Currently, it produces splits generally with num docs << 10M which is our general optimal num docs. This comes from the fact that we commit a split every 60 seconds (default commit_timeout_secs). This will lead to merges and will lower the indexing speed.

Ideally, we would need to produce split with num docs of 10M directly, this can be done by putting a high commit_timeout_secs by default.

I suggest putting the commit timeout to 3600 seconds to avoid merges.