Fix infer_type_of_indexed handling of MultiIndex

[seantalts]

Expr.Helpers.infer_type_of_indexed mis-types multi-indexing. vector[MultiIndex] returns real (it falls through to the scalar arm), matrix[MultiIndex] returns vector (it is grouped with the Single arms), and matrix[MultiIndex, Single] matches no arm and raises an internal error. The rule is that Single reduces a dimension while All, Upfrom, Between, and MultiIndex preserve it.

The bug is latent today because codegen ignores the type metadata. It matters for anything that trusts it: Memory_patterns.matrix_set stops recursing at non-Eigen types, so an expression wrongly typed real can leave a variable out of the SoA demotion set. I hit this while working on #1666, where a compiler-built gather typed real produced C++ that assigned an SoA value into an AoS slice and did not compile. Ast_to_Mir.copy_indices also feeds arbitrary index lists into this function and could hit the internal error.

The fix moves MultiIndex into the dimension-preserving arm and adds [MultiIndex, Single] to the matrix arms next to [All, Single]. New cases in the inline expect test cover vector, row vector, matrix, matrix-then-column, and arrays. No other test output changes.

Submission Checklist

• Run unit tests
• Documentation
  • OR, no user-facing changes were made

Release notes

Fixed the inferred type of multi-indexed expressions in the compiler's middle intermediate representation.

Copyright and Licensing

By submitting this pull request, the copyright holder is agreeing to
license the submitted work under the BSD 3-clause license (https://opensource.org/licenses/BSD-3-Clause)

I used AI but understand the code and think it makes sense.

[nhuurre]

Thanks. It seems lines 224, 226, and 227 implicitly rely on the wildcard index matching only Single _ because all others were already handled on line 209. (which was not the case for MultiIndex _ as you discovered)

I'd be more confident that this fixes all the corner cases if the final internal error arm was an exhaustive match instead of wildcard. I believe it should be

    | ( (UInt | UReal | UComplex | UTuple _ | UFun _ | UMathLibraryFunction)
      , _ :: _ )
     |(UVector | URowVector | UComplexVector | UComplexRowVector), _ :: _ :: _
     |(UMatrix | UComplexMatrix), _ :: _ :: _ :: _ ->
        ICE.(internal_errorf "Can't index %t" [UnsizedType.pp $ ut])
        [@coverage off]

[codecov]

Codecov Report

❌ Patch coverage is 59.09091% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.32%. Comparing base (b96c001) to head (99e8b21).
⚠️ Report is 26 commits behind head on master.

Files with missing lines	Patch %	Lines
src/middle/Expr.ml	59.09%	9 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1667   +/-   ##
=======================================
  Coverage   92.31%   92.32%           
=======================================
  Files          67       67           
  Lines        9972     9978    +6     
=======================================
+ Hits         9206     9212    +6     
  Misses        766      766           

Files with missing lines	Coverage Δ
src/middle/Expr.ml	84.21% <59.09%> (+0.74%)	⬆️

🚀 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.

[seantalts]

Done. Making it exhaustive also forced two arms the wildcard used to swallow: matrix with a Single row and a range column is a row_vector, and matrix with two range indices is a matrix. Both used to hit the internal error. Added test cases for those.

[nhuurre]

This looks like it covers both x[ : ,1] and x[1]. But x[1] is the same as x[1, : ]. I believe this is a pre-existing bug. Matrix[Single] should be a row vector.