coalesce outputs with default workable values

Add an optional --coalesce argument to relevant CLIs, with default values based on trade-offs between less output files and longer execution time.

Notes

• model.py was already using the default coalesce value of 8
• a drastic coalesce to the default value leads to crashes of Spark executors in embeddings.py, due to too few nodes handling the whole computation

Report

script	coalesce	files before	files after
sections.py	8	2049	9
embeddings.py	100	1025	101
features.py	4 ^	807k	1k

^ We used repartition, see https://phabricator.wikimedia.org/T350009#9389878.

Bug: T350009