Sprinkle @slow on slow tests

[A5rocks]

Fixes #3206.

The main thing is that it turns out that waiting for a socket to fail to connect takes up to 2 seconds on Windows -- so tests that rely on that should be marked @slow.

---

This approximately halves local test time on Windows.

================================================ slowest 10 durations =================================================
0.75s call     src/trio/_tests/test_exports.py::test_static_tool_sees_all_symbols[jedi-trio]
0.72s call     src/trio/_tests/tools/test_gen_exports.py::test_lint_failure
0.70s call     src/trio/_tests/test_repl.py::test_main_entrypoint
0.52s call     src/trio/_tests/test_util.py::test_coroutine_or_error
0.39s call     src/trio/_tests/test_exports.py::test_static_tool_sees_all_symbols[pylint-trio]
0.33s call     src/trio/_tests/test_subprocess.py::test_run
0.32s call     src/trio/_tests/test_subprocess.py::test_interactive[run_process in nursery]
0.32s call     src/trio/_tests/test_socket.py::test_many_sockets
0.26s call     src/trio/_tests/test_subprocess.py::test_interactive[open_process]
0.26s call     src/trio/_core/_tests/test_asyncgen.py::test_fallback_when_no_hook_claims_it
===================================== 762 passed, 83 skipped, 2 xfailed in 17.93s =====================================

[jakkdl]

2 seconds sounds unnecessarily slow, even if it's only CI. It'd be nice if we could lower the timeout somehow, or cancel it if it hasn't connected in a reasonable time. Though the former sounds difficult if there's no official API, and the latter would change what the test is actually testing for.

[A5rocks]

Based on the link in https://stackoverflow.com/a/54599972;

Upon receiving the ACK/RST client from the target host, the client determines that there is indeed no service listening there. In the Microsoft Winsock implementation of TCP, a pending connection will keep attempting to issue SYN packets until a maximum retry value is reached (set in the registry, this value defaults to 3 extra times)

It seems it only actually takes 1s, so we must be trying two ports (ipv4 and ipv6?).

Setting registry values doesn't seem like something a test suite should do locally.

[CoolCat467]

Looks pretty good, agree with jakkdl's comment as well though

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00000%. Comparing base (a9f8abd) to head (144ff95).
Report is 56 commits behind head on main.

Additional details and impacted files

@@                Coverage Diff                @@
##                 main        #3215     +/-   ##
=================================================
  Coverage   100.00000%   100.00000%             
=================================================
  Files             124          124             
  Lines           18792        20902   +2110     
  Branches         1268         1641    +373     
=================================================
+ Hits            18792        20902   +2110     

Files with missing lines	Coverage Δ
...c/trio/_tests/test_highlevel_open_tcp_listeners.py	100.00000% <100.00000%> (ø)
src/trio/_tests/test_socket.py	100.00000% <100.00000%> (ø)
src/trio/_tests/tools/test_gen_exports.py	100.00000% <100.00000%> (ø)

... and 20 files with indirect coverage changes