Fix send_all byte offsets for typed memoryviews

[AryanRG]

Fixes #3543.

SocketStream.send_all() and POSIX FdStream.send_all() treat the byte count returned by a partial write as a memoryview element offset. With a contiguous typed view, the retry can skip data and return successfully after sending only part of the buffer. A real-pipe reproduction on macOS sent 524,288 bytes from a 1,048,576-byte H view on unchanged main.

Cast nonempty views to a one-dimensional byte view before slicing. This also handles multidimensional and scalar views. Preserve empty strided-view handling and release the internal views without releasing the caller's view. A release note is included.

Validation on macOS:

• Six focused regression cases fail against base 4630d02872c0aa5778ce0afd702c5d532a2ccf60 and pass with the fix. Two empty-view compatibility cases pass in both versions. Tests cap real socket/pipe writes at three bytes for deterministic partial writes.
• Both affected modules: 26 passed on Python 3.10 and 3.12.
• Python 3.10 installed-wheel suite, including slow tests: 879 passed, 25 skipped, 2 xfailed.
• Mypy over all 147 source files passes for Darwin, Linux, and Windows targets. Applicable configured pre-commit hooks pass.
• Independent AI review: approved after an initial empty-strided-view regression was fixed and covered by tests.

Earlier full-suite attempts used the shell's Conda environment for Jedi or lacked the virtualenv's tools on PATH. The isolated wheel run above used the correct active virtualenv. Windows/Linux runtime execution remains for CI.

AI disclosure: this issue investigation, implementation, tests, and PR description were produced by an autonomous OpenAI Codex agent at the account owner's request. Validation and independent review were performed by AI agents; no human review is claimed.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00000%. Comparing base (4630d02) to head (1c8419e).

Additional details and impacted files

@@               Coverage Diff               @@
##                 main        #3544   +/-   ##
===============================================
  Coverage   100.00000%   100.00000%           
===============================================
  Files             128          128           
  Lines           19474        19529   +55     
  Branches         1323         1328    +5     
===============================================
+ Hits            19474        19529   +55     

Files with missing lines	Coverage Δ
src/trio/_highlevel_socket.py	100.00000% <100.00000%> (ø)
src/trio/_tests/test_highlevel_socket.py	100.00000% <100.00000%> (ø)
src/trio/_tests/test_unix_pipes.py	100.00000% <100.00000%> (ø)
src/trio/_unix_pipes.py	100.00000% <100.00000%> (ø)

🚀 New features to boost your workflow:

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

[A5rocks]

Hi! Thanks for the bug report and PR. I agree this is a bug, but I'm not willing to code review code someone else has not even reviewed. I think some parts of this PR is questionable (why not instantly cast? why not rely on truthiness instead of doing .nbytes?) so I'll close it.