Add support for "priority" import names

[jparise]

You can set the priority-import-names option to a comma separated
list of names that will be considered "priority" imports. These names
are allowed to appear before the standard library grouping, making it
easier to apply custom behavior (e.g. monkey patches) to subsequent
import statements.

For example:

import gevent.monkey
gevent.monkey.patch_all()

import socket

There doesn't appear to be another way to support this pattern. The error
would appear on the import socket line above, and that's not the correct
place to add a #noqa annotation.

I don't love the name "priority" for this. I also considered "early", but that's
equally ambiguous. I'm more than happy to rename this to something else.

[pgjones]

I'm not sure about this, given that none of the current styles specify a priority grouping. I'm also unsure as I think these is a rare use-case that actually might suit #noqa better. It may make more sense to restore # noqa on import gevent having an affect. Do you see any other use cases?

[jparise]

This is orthogonal to styles (akin to the __future__ prioritization), but regardless, I'm also in favor of the # noqa approach. Do you have a sense for what it would take to make that work for these cases?

[jparise]

#120 and #136 are good examples of the underlying use case.

[pgjones]

Don't both those examples show a situation in which you'd want to import a module first in one of the modules in the package likely followed by normally in the other modules?
For example I'd likely want to do,

# init.py
import eventlet
eventlet.monkey_patch_all()

import os

# other.py
import os

from eventlet import spawn

I'm still struggling to see the use case whereby you'd want a priority in every file.

[jparise]

I think you're right. In my gevent example, there's a distinct gevent.monkey module that is used for all monkey patching, but your eventlet example shows that's not the rule.

Let's pursue the # noqa approach (#136) instead.