Better handling if parent closes REDO_CHEATFDS or MAKEFLAGS fds.

Silently recover if REDO_CHEATFDS file descriptors are closed, because
they aren't completely essential and MAKEFLAGS-related warnings already
get printed if all file descriptors have been closed.

If MAKEFLAGS --jobserver-auth flags are closed, improve the error
message so that a) it's a normal error instead of an exception and b)
we link to documentation about why it happens.  Also write some more
detailed documentation about what's going on here.

t/204-makeflags/all.do

@@ -0,0 +1,16 @@
+# Make sure we can survive if a process closes all file descriptors,
+# including any jobserver file descriptors, as long as they also
+# unset MAKEFLAGS.
+redo-ifchange ../../redo/py
+
+# If we leave MAKEFLAGS set, then it's fair game to complain that the
+# advertised file descriptors are gone, because GNU make also complains.
+# (Although they only warn while we abort.  They can't abort so that
+# they don't break backward compat, but we have no such constraint, because
+# redo has always failed for that case.)
+#
+# On the other hand, we shouldn't have to unset REDO_CHEATFDS, both for
+# backward compatibility, and because REDO_CHEATFDS is undocumented.
+# redo should recover silently from that problem.
+unset MAKEFLAGS
+../../redo/py ./closefds.py redo noflags

t/204-makeflags/clean.do

@@ -0,0 +1 @@
+rm -f *~ .*~

t/204-makeflags/closefds.py

@@ -0,0 +1,12 @@
+import subprocess, sys, os
+
+# subprocess.call(close_fds=True) is unfortunately not a good idea,
+# because some versions (Debian's python version?) try to close inordinately
+# many file descriptors, like 0..1000000, which takes a very long time.
+#
+# We happen to know that redo doesn't need such huge fd values, so we'll
+# just cheat and use a smaller range.
+os.closerange(3, 1024)
+rv = subprocess.call(sys.argv[1:])
+sys.exit(rv)
+

t/204-makeflags/noflags.do

@@ -0,0 +1 @@
+true