fix leak in jeeps serial comm on posix.

[tsteven4]

This resolves #1221

I have tested this with a GPS12, "vtesto -l -j vg.log sgt.test", and:
gpsbabel -D3 -i garmin -f /dev/ttyS0
gpsbabel -D3 -i gpx -f w.gpx -o garmin -F /dev/ttyS0

both operations succeeded and no leaks were detected.

• This test was done on a hyper-v gen 1 ubuntu 22 guest running under a windows 11 host.
• The serial port was emulated through a USB to RS232 Converter Cable https://ftdichip.com/products/chipi-x10/.
• A windows named pipe was connected to the emulated serial port via https://github.com/tdhoward/COMpipe with the command "COMpipe.exe -b 9600 -c \\.\COM3 -p \\.\pipe\gen1ubuntu22".
• The hyper-v guest settings set guest COM 1 to the named pipe gen1ubuntu22.

[tsteven4]

thanks @tdhoward for COMpipe.

[robertlipe]

I used to build serial drivers (and hardware) professionally. We were dealing with things that werent really a 16450 running on a Real Computer but virtualizing down to a 16450 in DOS so that Procomm and Kermit and things that wanted to do outb()'s on 0x3f8 would actually go through ISA shared memory commands, be read by one processor that packetetized that and generated sync packets, then shipped them over potentially a modem or even an SDLC connection, then get picked off the wire and command (alien) uarts on the other side of all that to do their bidding. It was a beautiful series of hacks when it worked ... and a hell of a train wreck when it didn't.

So of course I approve the PR, but I have a warm feeling about that glorious testing process. Nicely done!