Publish nightly builds to halide/pypi instead of pypi.halide-lang.org

[alexreinking]

Summary

• Replaces the push-to-main nightly publish step (pypa/gh-action-pypi-publish targeting pypi.halide-lang.org) with gh release create/upload against halide/pypi, tagged halide@<version>, plus an explicit index-rebuild trigger. The tagged-release publish step below it (real releases -> pypi.org via OIDC trusted publishing) is untouched.
• Removes the retry/diagnostics wrappers (5 attempts, 60s->300s backoff, egress IP + traceroute logging) around uv sync/uv pip install in testing-linux.yml, testing-arm-linux.yml, testing-macos.yml, and testing-windows.yml. Those existed specifically to work around MIT's border-security IP quarantines on the old pypiserver VM; now that pypi.halide-lang.org serves halide/pypi (GitHub Releases + Pages), which CI already trusts implicitly for everything else, that failure mode no longer applies.
• Same publish pattern already proven out on halide-llvm (merged, Publish wheels to halide/pypi instead of pypi.halide-lang.org halide-llvm#1) and halide-wheel-deps.

Why

pypi.halide-lang.org now serves halide/pypi instead of the retired MIT-hosted pypiserver, for the same reason as the halide-llvm migration: that VM's network path was subject to intermittent border-security IP quarantines outside anyone's control. This was deliberately held back from the earlier halide-llvm migration to validate the consumption side first; that's now confirmed working end-to-end (real halide-llvm wheels built, published, and merged), so this closes the loop for Halide's own nightly builds and removes the now-dead workaround code.

Test plan

• actionlint passes on all five modified workflow files.
• Confirmed halide's wheel filenames start with halide-<version>-... (no normalization surprises, matches dist/halide-*.whl cleanly).
• Confirmed the only remaining "retry" hits in testing-linux.yml/testing-arm-linux.yml are an unrelated apt-get update retry loop, correctly left alone.
• Publish step not yet exercised for real (nightly-only trigger) -- recommend merging and watching the next main push's publish job.

[derek-gerstmann]

LGTM.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.28%. Comparing base (0dba9f1) to head (24b7c14).
⚠️ Report is 6 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #9282      +/-   ##
==========================================
+ Coverage   70.14%   70.28%   +0.14%     
==========================================
  Files         257      258       +1     
  Lines       79162    79162              
  Branches    18979    18997      +18     
==========================================
+ Hits        55529    55642     +113     
+ Misses      17881    17799      -82     
+ Partials     5752     5721      -31     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.