See my comment on the other review.

I think isolation by default is good, because:

1. Isolation is generally good. I think @dotordogh or I ran into issues where

   pants/tests/python/pants_test/engine/test_fs.py

   Lines 406 to 423 in db713e4

   	def test_materialize_directories(self):
   	# I tried passing in the digest of a file, but it didn't make it to the
   	# rust code due to all of the checks we have in place (which is probably a good thing).
   	self.prime_store_with_roland_digest()
   	with temporary_dir() as temp_dir:
   	dir_path = os.path.join(temp_dir, "containing_roland")
   	digest = Digest(
   	text_type("63949aa823baf765eff07b946050d76ec0033144c785a94d3ebd82baa931cd16"),
   	80
   	)
   	scheduler = self.mk_scheduler(rules=create_fs_rules())
   	scheduler.materialize_directories((DirectoryToMaterialize(text_type(dir_path), digest),))
   	created_file = os.path.join(dir_path, "roland")
   	with open(created_file, 'r') as f:
   	content = f.read()
   	self.assertEqual(content, "European Burmese")

   was either failing when it should pass, or passing when it should fail, before the prime_store_with_roland_digest was added, because it relied on the store already happening to have the digest in question in it.
2. The speedup of re-using an existing store should be pretty small.

(The "force isolation" function is handy because the default in this PR is actually to share a store across an test class)

Happy to re-visit if you are convinced otherwise :)