stdinit initializes the first three slots of the files array
with statically allocated pathnames. closefile considers these
slots permanently statically allocated. openfile does not.

This discrepancy can leak memory: if a statically allocated slot
is vacated, openfile will put a malloc'd string where closefile
does not free.

To reconcile, either openfile must not reuse these slots or
stdinit and closefile must switch to dynamic allocation. I chose
the latter, because the result is a little simpler and less
interconnected than managing a partitioned array.

Here's a (contrived) demo with my system's output:

    $ cat leak.sh
    #!/bin/sh

    # Close stderr, vacating files[2]
    # Print the initial resident set size
    # For each record, open and close a file named "y"
    # Print rsz every 50,000 iterations
    # Exit after 250,000 iterations

    awk=${1:-nawk}
    yes | $awk '
            BEGIN {
		close("/dev/stderr")
		system(rsz = "ps -p $PPID -o rsz=")
	    }
	    { print > $0; close($0) }
	    NR % 50000 == 0 { system(rsz) }
	    NR == 250000    { exit }'

    # End of leak.sh

    $ paste <(./leak.sh ./master) <(./leak.sh ./a.out)
     2444	 2428
     4016	 2500
     5576	 2500
     7140	 2500
     8704	 2500
    10264	 2500