[blegat]

Needs this one to be merged first, I'll rebase afterwards (so at the moment the diff contains both):

• Speed up CSC substitution decompression for stars #354

Add @inbounds to the CSC recovery loops and propagate it through the shared write helper with Base.@propagate_inbounds.

Fresh median timings for the same benchmark as #354:

Implementation	Recovery only	Full Hessian
Legacy MOI	98.07 μs	507.54 μs
SMC main	243.47 μs	666.90 μs
#354	137.98 μs	520.81 μs
This PR	88.59 μs	502.87 μs

On Julia 1.13.1 with Intel i7-12700K.

[codecov]

Codecov Report

❌ Patch coverage is 2.56410% with 38 lines in your changes missing coverage. Please review.
✅ Project coverage is 46.73%. Comparing base (7d894ca) to head (544fe40).

Files with missing lines	Patch %	Lines
src/decompression.jl	0.00%	38 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #355      +/-   ##
==========================================
- Coverage   46.91%   46.73%   -0.19%     
==========================================
  Files          20       20              
  Lines        2498     2508      +10     
==========================================
  Hits         1172     1172              
- Misses       1326     1336      +10     

☔ 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.

[gdalle]

We've had this discussion with @amontoison in the past and I'm very uncomfortable adding @inbounds to a loop without previously checking that all indexing falls in the proper range. However I think this could be an opportunity to tighten our typing, and maybe put Vectors everywhere we can, to simplify this kind of verification (there are lots of places where AbstractVectors are accepted but really shouldn't be).