[mlakolar]

Summary

Widens the CoordinateDescent compat to "0.4, 0.5" (version 0.2.1) and adapts the package to the two breaking changes of CoordinateDescent 0.5 that affect it.

• Options in the second stages. Each estimator ran a second stage with options rebuilt field by field (CDOptions(; maxIter=options.maxIter, …, warmStart=true)). That drops relativeTol, the field CoordinateDescent 0.5 adds, so a caller's relativeTol = false would be ignored. _warmStart(options) copies every field and sets warmStart, and works with both versions.
• Relative stopping rule. CoordinateDescent 0.5 stops relative to coordinateScale and responseScale of the loss, and losses that do not define them keep the absolute rule. CDInverseKroneckerLoss and CDInverseReducedSymKroneckerLoss now define them as CDQuadraticLoss does: the square root of the curvature along each coordinate, and the largest |bₖ| / √aₖ. The definitions are guarded by isdefined(CoordinateDescent, :coordinateScale), so the package still loads with 0.4. This matches CovSel's change for CDDirectDifferenceLoss (Allow CoordinateDescent 0.5 CovarianceSelection.jl#8).

No API changes.

Tests

• _warmStart keeps every field, including relativeTol
• with CoordinateDescent 0.5: the scales of both losses equal those of CDQuadraticLoss with the same Hessian and linear term, and multiplying Σx and Σy by a power of 2 divides the solution by its square and takes the same number of passes
• all tests pass locally on Julia 1.10 and 1.13 with CoordinateDescent 0.5.1 + ProximalBase 0.4.0 (20184 tests) and with CoordinateDescent 0.4.1 + ProximalBase 0.3.2 (20164), against CovSel master with its compat widened as in README: the five packages in one table, near the top #8

Merge order

CI takes CovSel from its GitHub master, which still requires CoordinateDescent 0.4 until mlakolar/CovarianceSelection.jl#8 is merged, so CI here tests 0.4 only. Merging this first is safe; rerun CI after CovSel#8 is merged to test 0.5.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.64%. Comparing base (d88f552) to head (0cc0f48).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master       #4   +/-   ##
=======================================
  Coverage   98.64%   98.64%           
=======================================
  Files          12       12           
  Lines        1403     1407    +4     
=======================================
+ Hits         1384     1388    +4     
  Misses         19       19           

☔ View full report in Codecov by Harness.
📢 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.