gh-94609: Make test_ssl.ThreadedEchoServer exceptions appear in a main thread

[arhadthedev]

Feature or enhancement

This PR:

• adds test.test_ssl.Server that is like test.test_ssl.ThreadedEchoServer but (1) implemented as a context manager and (2) replaces plain printing of server-side tracebacks with bubbling the exception up between threads through a future. The (2) is implemented as follows:
  • the server force-closes a client connection when exceptions occur so a client thread instantly gets ConnectionAbortedError / ConnectionResetError hence knowing that the future is going to have a server exception inside
  • the client-side context manager gets the connection failure and rethrows the prepared server exception from the future
• converts some test cases into using the new approach; others will be converted later in separate PRs
• enforses a timeout for all socket operations using setUp/tearDown
• teaches tests to push empty packets because TLS 1.3 delays the handshake until some data are sent

Pitch

• expected exceptions (caused by *_fail test cases) do not flood a console with unnecessary stack traces (and a heart of a watcher with fear)
• instead, they can be captured by assertRaises in a main thread; unexpected exceptions will migrate into a Tests Result report footnote failing the run so no need to scroll the log hundreds lines up
• a server-side exception won't hang a main thread until a socket reading timeout happens
• replacement of test_ssl and test_asyncio threaded servers with a unified implementation (they cannot be ported to asyncio because a client part is blocking and requires thread separation anyway)

TODO

Plans for future PRs:

• port all test cases to the new class
• remove the original ThreadedEchoServer
• port other test suits that use multithreaded servers (including the ones that currently use asyncore like test_ftp and test_logging)

• Issue: test_ssl is too verbose #94609

[arhadthedev]

A thread pool triggers ENV CHANGED. Any ideas how to remedy this without creating&destroying a loop for each individual test?

== Tests result: ENV CHANGED ==
425 tests OK.
10 slowest tests:
- test_tools: 5 min 45 sec
- test_concurrent_futures: 2 min 54 sec
- test_multiprocessing_spawn: 2 min 40 sec
- test_gdb: 2 min 7 sec
- test_multiprocessing_forkserver: 1 min 28 sec
- test_asyncio: 1 min 27 sec
- test_venv: 1 min 17 sec
- test_multiprocessing_fork: 1 min 17 sec
- test_signal: 1 min 5 sec
- test_regrtest: 50.5 sec
1 test altered the execution environment:
    test_ssl
10 tests skipped:
    test_devpoll test_ioctl test_kqueue test_launcher test_msilib
    test_startfile test_winconsoleio test_winreg test_winsound
    test_zipfile64
Total duration: 11 min 1 sec
Tests result: ENV CHANGED
make: *** [Makefile:1863: buildbottest] Error 3
Error: Process completed with exit code 2.

[arhadthedev]

Any ideas how to remedy this without creating&destroying a loop for each individual test?

Not relevant anymore. The creationa and destruction happens for some modules only.

Whatever, it's ready for a review.

[arhadthedev]

Everything is green and ready for a review (after the release blocker crunch is over, of course).

[arhadthedev]

@tiran Would you mind to look at this PR? It adds a unified separate-threaded test.support.threading_helper.Server that I'm planning to use as a foundation to replace asynchat-based servers where asyncio is not applicable.

[arhadthedev]

I've added init() to avoid a modified environment test failure due to a leaked object. If there's a better solution, I'll use that instead.