Fix #3131: javalib ServerSocket should now be more accepting

[LeeTibbert]

Fix #3131

javalib java.net.ServerSocket#accept now accepts new connections without
the cast errors.

Collaboration

User devlam, the author of PR #3131, deserves "co-authored-by" recognition for the
precise localization & reproducing code provided in that PR.

Of course, all bugs remain "authored-by" user LeeTibbert.

Backport

This PR is a candidate for backport to the SN 0.4.x series. I believe/hope it
will backport cleanly. I checked the dates for the IPv6 backport to the SN 0.4.x series and
these changes are relevant to 0.4.x.

Later: I have not been able to get a private baseline (pre-changes of this PR) SN 0.4.x out-of-the-box
Scala 3.2.2, Java 8 build to work after several hours invested. I'll try again if/when my patience recovers.

Given that network code is notorious for being difficult to backport, I will be
trying a private build whilst this PR is in CI.

Later yet: I got a baseline SN 0.4.x out-of-the-box build to work and tried to apply
the` fixes in this PR. Using only the one file with the effective change
caused compilation errors due to differences in 0.4.x.

               I think the best thing for me to do is create a separate 0.4.x PR.
               I already have the minimal change and have tested it.  No IPv6
               on SN 0.4.n, so testing takes less time.  I want to create the
               PR at the beginning, not the end, of a sprint.

Testing:

This PR has been manually tested with both IPv6 & IPv4 connections and passes.
scripted-test java-net-server-socket is intended to exercise accept() but appears
totally non-functional; see Issue #3139.

[WojciechMazur]

Looks good! Don't worry about spurious CI failures involving multithreading / Commix, the fix for them is on its way.

[LeeTibbert]

Wojciech, Thank you for the timely review and encouragement.

I was shamelessly punting the multithread errors, figuring that effort is
not complete yet.

I will continue to monitor this CI for what may be "my" errors.

I am at end-of-sprint now but will try a baseline 0.4.x build tomorrow-ish.
If successful, I will then add the changes of this PR. Given all
you went thru, twice, with NetworkInterface.scala, I will feel better
knowing that this backports.

[LeeTibbert]

For those who may be looking at this PR in the future:

The effective change is in AbstractPlainSocket.scala, around the inet_mumble() code.

I large number of other changes are an attempt to consistently use the "at1.at(0)". in
the sin6_addr case. "at1" works because it references the same location as "at1.at(0)".
The "at1.at(0)" idiom is longer to read and more typing to write, but it gives a better
correspondence of types: that is, a Ptr[Byte] rather than a Ptr[CStruct1].