[BUG] Stop embedded HTTP server writes from raising SIGPIPE

[thc1006]

Fixes #4293
Refs #4287

Opened as a draft while CI runs. Ready for review from my side.

The problem

Socket::send() passed a flags argument of 0, and there was no MSG_NOSIGNAL or SO_NOSIGPIPE anywhere under ext/. POSIX raises SIGPIPE on a write to a connected socket that can no longer transmit, in addition to failing with EPIPE, and its default disposition terminates the process. The embedded server could therefore be killed before any of the error handling in sendMore() ran, which also means fixing the fatal send error item in #4287 on its own would not have been enough.

Checked through SocketTools::Socket::send() itself, on a socketpair whose peer end is closed, forking so the signal is observable:

=== before ===
  Socket::send -> KILLED by signal 13 (Broken pipe)

=== after ===
  Socket::send -> process exited 0

The fix

• send() passes MSG_NOSIGNAL on platforms that define it, which covers Linux.
• Platforms that offer SO_NOSIGPIPE instead, which is macOS and the BSDs, get the socket option set once per socket: in the (af, type, proto) constructor and on the socket handed back by accept().
• Windows needs neither and compiles both branches out.

The suppression is per socket on purpose. Installing a process-wide SIGPIPE handler or setting it to ignored would work too, but that is a decision belonging to the program embedding this server, not to a header it includes.

On testing

The check above forks and inspects the wait status, which does not translate cleanly into a normal gtest assertion. The gtest form would be a death test, and those are restricted under some sanitizers and behave differently on Windows.

There is also no home for it yet: #4292 introduces ext/test/http/socket_tools_test.cc, and adding the same file here would guarantee a conflict between the two. I would rather add this as a follow-up on that target once #4292 lands, and I am happy to do that. If you would prefer the test in this PR instead, say so and I will rebase this on top of #4292.

Verification

• clang-format 18 clean.
• clang-tidy 18 with the repo .clang-tidy on socket_tools.h: 2 warnings before and 2 after, both pre-existing. warning_limit is untouched.
• MSG_NOSIGNAL and SO_NOSIGPIPE are both used behind #ifdef, so no platform sees an undefined identifier.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.24%. Comparing base (d3c7a47) to head (39ef641).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4294      +/-   ##
==========================================
- Coverage   81.31%   81.24%   -0.06%     
==========================================
  Files         445      446       +1     
  Lines       18859    18875      +16     
==========================================
  Hits        15333    15333              
- Misses       3526     3542      +16     

Files with missing lines	Coverage Δ
...clude/opentelemetry/ext/http/server/socket_tools.h	94.12% <100.00%> (+0.12%)	⬆️

... and 18 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.