Optimize pure delete lock updates.

[jsirois]

When a lock update only asks for deletes, we now process the "resolve"
offline without consulting Pip.

Fixes #2566

[jsirois]

N.B.: Existing tests cover all the interesting delete cases, including transitive dependencies of deleted node being partially retained when something else depends on those, and they pass.

Here proj-b 1 depends on proj-a and proj-c:

pex/tests/integration/cli/commands/test_lock_update_issues_2332_2334.py

Lines 441 to 464 in e0aac4f

	def test_lock_update_mixed(
	run_lock_update, # type: RunLockUpdate
	lock, # type: str
	path_mappings, # type: PathMappings
	):
	# type: (...) -> None
	run_lock_update("-p", "proj-a", "-d", "proj-b").assert_success(
	expected_error_re=re_exact(
	dedent(
	"""\
	Updates for lock generated by universal:
	Deleted proj-b 1
	Deleted proj-c 1
	Updated proj-a from 1.1 to 2
	Updates to lock input requirements:
	Deleted 'proj-b==1.*'
	"""
	)
	)
	)
	lockfile = json_codec.load(lock, path_mappings)
	locked_requirements = locked_requirement_pins(lockfile)
	assert {pin("proj-a", "2"), pin("ansicolors", "1.1.8")} == locked_requirements

[benjyw]

Nice!

[jsirois]

My claim of an existing complex delete case in #2569 (comment) was a lie, only simple deletes are tested through the new code path. That case has a delete and an update; so does not use the new code path and uses Pip instead. I've added a complex pure delete test case in #2576. It passes with no changes; so I'll process with the release before landing that change.