Make prefix more type-stable

[penelopeysm]

While trying to hunt down type instabilities in TuringLang/DynamicPPL.jl#892 this was one of the things that I first looked at. I thought that the changes in this PR would make a difference. However, simply testing @inferred AbstractPPL.prefix(vn1, vn2) with various combinations of vn1 and vn2, I can't actually see any difference on Julia 1.11.5, and it didn't help with any upstream type instabilities.

It appears to me that the compiler is smart enough to figure out that getoptic(vn) returns the second type parameter of vn, and thus when it looks at the line if getoptic(vn) == identity it can statically determine whether this is true or false, and infer the return type appropriately.

I don't know if it's risky to rely on this behaviour, hence why I'm keeping this PR open.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.02%. Comparing base (87e3120) to head (88e9a95).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #121      +/-   ##
==========================================
+ Coverage   90.93%   91.02%   +0.09%     
==========================================
  Files           2        2              
  Lines         298      301       +3     
==========================================
+ Hits          271      274       +3     
  Misses         27       27              

☔ View full report in Codecov by Sentry.
📢 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.

[coveralls]

Pull Request Test Coverage Report for Build 14550376651

Details

• 11 of 11 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.09%) to 91.03%

---

Totals
Change from base Build 14321492427:	0.09%
Covered Lines:	274
Relevant Lines:	301

---

💛 - Coveralls

[yebai]

It's probably always a good idea to make the compiler's life easier where possible.

[sunxd3]

it looks fine to me and I don't see a downside for this over the original implementation.

Regarding fixing type stability errors, I has couple of thoughts, none of them are too serious:

1. It appears to me that the compiler is smart enough to figure out that getoptic(vn) returns the second type parameter of vn, and thus when it looks at the line if getoptic(vn) == identity it can statically determine whether this is true or false, and infer the return type appropriately.

   is it possible to test the behavior of the compiler in some way? (I am thinking inspecting IRCode)

2. (I don't have a good enough knowledge of the Julia compiler.) Topic_prefix is nested, right? Do we have an examples of reasonably deeply nested VarName? So that we can know that the compiler doesn't give up too easily?