Index send_all memoryviews by byte count, not element count

[Eryx-Z]

Summary

SocketStream.send_all and POSIX FdStream.send_all used byte counts from send() / os.write() as memoryview element offsets. For a contiguous view with itemsize > 1, a partial write could skip data and return successfully without sending the full buffer.

Cast nonempty contiguous views to a one-dimensional bytes view before slicing so offsets and lengths stay in bytes.

Fixes #3543.

Test plan

• New FdStream regression with memoryview(...).cast("H") over a pipe
• Existing test_SocketStream_send_all / pipe send tests still pass

[codecov]

Codecov Report

❌ Patch coverage is 61.90476% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 99.95896%. Comparing base (4630d02) to head (0e9a80c).

Files with missing lines	Patch %	Lines
src/trio/_tests/test_unix_pipes.py	56.25000%	7 Missing ⚠️
src/trio/_highlevel_socket.py	50.00000%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@                 Coverage Diff                  @@
##                 main       #3545         +/-   ##
====================================================
- Coverage   100.00000%   99.95896%   -0.04104%     
====================================================
  Files             128         128                 
  Lines           19474       19494         +20     
  Branches         1323        1326          +3     
====================================================
+ Hits            19474       19486         +12     
- Misses              0           7          +7     
- Partials            0           1          +1     

Files with missing lines	Coverage Δ
src/trio/_unix_pipes.py	100.00000% <100.00000%> (ø)
src/trio/_highlevel_socket.py	99.15966% <50.00000%> (-0.84034%)	⬇️
src/trio/_tests/test_unix_pipes.py	95.73171% <56.25000%> (-4.26829%)	⬇️

🚀 New features to boost your workflow:

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

[A5rocks]

Same as the other PR, I'm not willing to guide an LLM step by step into contributing. I can do that myself!

(and I don't really get why you'd want to special case contiguous memory views...?)