Skip the redundant permute in matricize, returning a view for dense arrays

[mtfishman]

Summary

matricize/matricizeop now skip the permuted copy when the requested row and column grouping is already in storage order, instead of always permuting first. A new matricizekind classifier, dispatched on FusionStyle, decides this per bipermutation: the generic classifier recognizes the always-safe aligned case (skipping a no-op permute is valid for any style), and ReshapeFusion (dense) additionally recognizes a pure codomain/domain swap, which it realizes as a lazy transpose. For a dense array the aligned and swapped cases return a reshape/transpose view of the input. For a graded array matricize still gathers blocks into a new matrix, but the redundant permute copy beforehand is skipped. The fast paths require op === identity, since a plain view cannot carry a fused op like conj. The result may alias the input and is read-only, which matches the matricizeop docstring's existing contract.

This removes input-copy allocations from aligned contractions. At bond dimension 64 the dense matmul runs about 35% faster and a memory-bound rank-3 contraction about 50% faster, closing most of the gap to an optimized reference, and an aligned AbelianGradedArray contraction drops one operand's permute copy.

The contraction driver still allocates the product matrix between the matrix multiply and the destination, and the factorizations still copy their input. Removing the product temp and adding an ownership-preserving matricize_copy will land in follow-up PRs.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.07%. Comparing base (289b6c7) to head (0fc13e3).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #183      +/-   ##
==========================================
+ Coverage   78.84%   79.07%   +0.23%     
==========================================
  Files          20       20              
  Lines         657      669      +12     
==========================================
+ Hits          518      529      +11     
- Misses        139      140       +1     

Flag	Coverage Δ
docs	30.20% <63.63%> (+0.37%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.