Fix TypeError for nonsquare linear systems with QR factorization

[ChrisRackauckas-Claude]

Summary

• Fixes TypeError: in setfield! for nonsquare linear system #674 - TypeError when solving nonsquare linear systems
• Ensures QRPivoted pivot indices are always Int64, regardless of platform

Problem

When solving nonsquare linear systems (A*u = b where A is m×n with m≠n), some platforms return QRPivoted with Vector{Int32} for pivot indices from Julia's qr function with ColumnNorm pivot, while LinearSolve's DefaultLinearSolverInit struct expects Vector{Int64}.

Solution

Added a helper function _ensure_int64_pivot that converts QRPivoted pivot vectors from Int32 to Int64 when necessary, ensuring compatibility across all platforms.

Changes

• Added _ensure_int64_pivot helper function to convert pivot indices to Int64
• Applied conversion in do_factorization for QRFactorization
• Applied conversion in all init_cacheval functions that return QRPivoted
• Ensured PREALLOCATED_QR_ColumnNorm uses Int64 pivot indices

Test Plan

• Created test case reproducing the original issue from TypeError: in setfield! for nonsquare linear system #674
• Added test for Int32 to Int64 pivot conversion
• Tested with various nonsquare matrix sizes (m×n where mn)
• All new tests pass locally
• Formatted code with JuliaFormatter using SciMLStyle

Test Results

Test Summary:                                   | Pass  Total  Time
Issue #674 - NonSquare QR Int32/Int64 pivot fix |   13     13  0.2s
All tests passed\!

🤖 Generated with Claude Code