fix reversion in recent CoW changes

File handle accidentially left open is both a FD leak and causes the
haskell RTS to reject opening it again with "file is locked".

doc/bugs/fresh_3_tests_fails-_openBinaryFile__58___resource_busy.mdwn

@@ -23,3 +23,5 @@ looked only at "normal" run
 so  actually looks like the same test in various scenarios.
 
 probably relates to the tune ups to make importtree work with CoW
+
+>  [[fixed|done]] --[[Joey]] 

doc/bugs/fresh_3_tests_fails-_openBinaryFile__58___resource_busy/comment_1_d78cd2a78c97a00629aa86b46e72ff67._comment

@@ -0,0 +1,7 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 1"""
+ date="2021-04-20T15:10:59Z"
+ content="""
+Indeed it was, file handle left open. Fixed.
+"""]]

doc/bugs/tests_fail_on_windows__58___retrieveKeyFile_resume/comment_1_59d56dcb2441f6c69b98605498460896._comment

@@ -0,0 +1,17 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 1"""
+ date="2021-04-20T15:10:00Z"
+ content="""
+It's possible it had the same cause as the other failure, which I've now
+fixed. That involved a file handle leak, and on windows a file handle being
+left open for write is treated the same as the file being locked and will
+prevent a later deletion attempt and also a later write attempt, which
+could explain the two unlike failures.
+
+However, the file that was left open was a temp file in the remote,
+not the object file in the annex. So I'm not sure if it's fixed.
+Also possible something else in the windows code path changed accidentially
+during those CoW changes. Will have to see what happens when the
+autobuilder runs again.
+"""]]