Changelog edit and fuzz for #139

[finnbear]

Followup to #139

• Changelog
• Fuzz (make sure ordering of new algorithm is correct)

Noticed #140 (infinity this time, not NaN), which needs a new fix, but that's orthogonal (only affects near-miss rays, so I simply restrict the new assertions to Mode::Grid)

^{I intend to wait ~24h before merging, just in case there are any comments/suggestions/objections.}

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 0% with 30 lines in your changes missing coverage. Please review.

Project coverage is 48.72%. Comparing base (9e6b312) to head (3a0dd2b).

Files with missing lines	Patch %	Lines
fuzz/fuzz_targets/fuzz.rs	0.00%	30 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #143      +/-   ##
==========================================
- Coverage   49.81%   48.72%   -1.09%     
==========================================
  Files          17       17              
  Lines        1347     1377      +30     
==========================================
  Hits          671      671              
- Misses        676      706      +30     

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

[svenstaro]

I'm not sure. I think it's good for the ecosystem to always run the latest nightly. We will otherwise definitely forget to upgrade. Hard regressions usually get fixed quickly.

[finnbear]

I think it's good for the ecosystem to always run the latest nightly.

Yes, it is possible to sacrifice reliability on our end to increase the chance that we find (and report?) issues to upstream rustc for the benefit of the ecosystem.

Never upgrading is also a bad policy, as eventual developments in locally-installed nightly Rust will produce code that doesn't work on the CI's version.

I personally view the harms of rustc-roulette to be greater, as it creates collateral damage to unrelated PR's. Never upgrading only risks errors when updating the fuzzer itself.

A real solution might involve trying the latest and producing a warning while falling back to a known good version if it doesn't work, but it seems complex to implement in CI.

I'll merge, but I'll make a tracking issue to help avoid forgetting about this.

[svenstaro]

Works for me!