Fix Zinc worker performance regression

[arturaz]

Fixes the delay when using Mill RPC infrastructure and sending large messages over it.

The cause was PipedInput/OutputStream waiting for one second when it's internal buffer fills up. Because ProxyStream protocol used a very small (126 byte) buffer, things got slow very fast once a large message was sent.

This PR removes the dependence on the piped streams and uses socket streams directly. It also ditches ProxyStream protocol for Mill RPC and instead sends heartbeats as empty lines.

Other changes:

• Server now is basic server, ProxyStream specific logic was moved to ProxyStreamServer.
• A lot more logging is now done. You can find these logs in out/mill/javalib/JvmWorkerModule/internalWorker.dest/jvm-worker.log and out/mill/javalib/JvmWorkerModule/internalWorker.dest/zinc-worker/DAEMON_HASH/daemon/server.log respectively.
• To aid debugging DebuggingInputStream and DebuggingOutputStream was created and left in the codebase if we'll need to debug something in the future.
• mill.util.Timed was introduced as a utility to time code blocks.

This has been tested manually:

arturas@tuxedo-ibp14-gen9:~/work/mill-javalib-1-simple$ rm -rf out* in_* && time ../mill/mill-assembly.jar foo.compile
[build.mill-59/64] compile
[build.mill-59] [info] compiling 3 Scala sources to /home/arturas/work/mill-javalib-1-simple/out/mill-build/compile.dest/classes ...
[build.mill-59] [info] done compiling
[35/35] foo.compile
[35] [info] compiling 1 Java source to /home/arturas/work/mill-javalib-1-simple/out/foo/compile.dest/classes ...
[35] [info] done compiling
[35/35] ============================== foo.compile ============================== 8s

real    0m10,160s
user    0m5,591s
sys     0m0,614s
arturas@tuxedo-ibp14-gen9:~/work/mill-javalib-1-simple$ time ../mill/mill-assembly.jar clean
[1/1] clean
[1/1] ============================== clean ==============================

real    0m0,266s
user    0m0,288s
sys     0m0,071s
arturas@tuxedo-ibp14-gen9:~/work/mill-javalib-1-simple$ time ../mill/mill-assembly.jar foo.compile
[35/35] foo.compile
[35] [info] compiling 1 Java source to /home/arturas/work/mill-javalib-1-simple/out/foo/compile.dest/classes ...
[35] [info] done compiling
[35/35] ============================== foo.compile ============================== 1s

real    0m1,470s
user    0m0,387s
sys     0m0,109s

The first run has to spawn a Zinc subprocess, so it takes 10s, but then the subsequent compilations are fast, as we expect them to be.

---

Fixes #5693

[arturaz]

@lihaoyi does this look sensible, silencing the exception in case the thread was interruped?

[lihaoyi]

Not sure, does this happen regularly? If so skipping the logging seems fine

[arturaz]

I've noticed it when running netty tests, I was getting spammed in the console with random looking stacktraces, which has lead to 14a2986. So pretty regularly in bigger builds.