Add more bibtex tests

[CraftSpider]

These test some simple error cases, I plan to rebase #1032 on this so there are more tests available to avoid regressions

[codecov]

Codecov Report

Merging #1037 (a6b87ba) into master (0d9169e) will increase coverage by 0.06%.
The diff coverage is 76.19%.

❗ Current head a6b87ba differs from pull request most recent head 84ec77c. Consider uploading reports for the commit 84ec77c to get more accurate results

@@            Coverage Diff             @@
##           master    #1037      +/-   ##
==========================================
+ Coverage   44.79%   44.85%   +0.06%     
==========================================
  Files         155      155              
  Lines       62607    62585      -22     
==========================================
+ Hits        28043    28075      +32     
+ Misses      34564    34510      -54     

Impacted Files	Coverage Δ
src/config.rs	60.52% <0.00%> (ø)
crates/engine_bibtex/bibtex/bibtex.c	44.75% <33.33%> (+1.40%)	⬆️
src/driver.rs	75.65% <100.00%> (+0.11%)	⬆️

... and 3 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[CraftSpider]

Well, apparently, the existing code fails on Windows. It doesn't fail locally on Linux or Windows, which makes me think somehow the code is invoking UB, except I think it's UB in the C code, so no clue where it is yet. I'll spend some time trying to figure this out.

[CraftSpider]

Ah, it's not UB, I realize what's wrong I think - \r and \n are considered separate newline characters on Windows, and I bet it's checking out with \r\n newlines - that would explain why the numbers are what they are, it's counting each line twice. I think I can fix that in input_ln

[pkgw]

Thanks for working on this! I have a couple of requested changes that ought to be straightforward to implement, if you don't mind.

[pkgw]

Thanks! Looks like we had a spurious test failure here, I think that we should pass with a retry.

(The tectonic-on-arXiv test will just raise some failures incorrectly for a bit until we straighten out some reproducibility subtleties; see #1036)