The workflow is directly inspired by git-as-a-wrapper-around-Perforce used at Google.
It's nice to have a cohesive unit of change in a single commit, but if your mergeable features have a lot of moving parts it may be too much information in one giant blob, losing valuable history. There are certain things I like to be in individual commits, like for instance, DB migrations in Rails or Django, or Gemfile/Gemfile.lock upgrades. Having these in separate commits allows for easier conflict resolution later, and are also a nice granularity for practices like zero-downtime continuous deployment where you need to have multi-phase deploy of migrations (ie. phase 1: deploy migrations that add fields, phase 2: deploy code that requires new fields, phase 3: remove fields required by legacy code).
Maybe in a large team with a large volume of PRs then squashing by default is preferred so you can make sense of a fast-moving history, but given that it's so easy to squash selectively with `git rebase -i` that is my preferred method, at least for the team I'm on now.
A lot of devs see commit messages and rebasing as useless pedantry, but once you start to take it seriously it only takes a little bit of extra time, and it provides so much future benefit via git-bisect and general spelunkability.
I also think it's worth investing more time in a tool based on its longevity in ones workload. In that regard I put git at the top of the food chain, as I expect to use it even longer than Linux, for example in my case: Git > Linux > Vim > Ruby/Python > OS X / GUI tools.
A side effect of being this good at git is the ability to bail out your colleagues when they royally fuck up a merge or rebase.