Skip the zeroed-destination gather in matricized contraction

[mtfishman]

Summary

When β is a strong zero, contractopadd! for the Matricize algorithm no longer matricizes a_dest to form the mul! target. Matricizing the destination gathers all of its (about-to-be-overwritten) blocks into a matrix that mul! then immediately overwrites, which for a graded output is a full block-by-block copy. Instead the matmul allocates its result directly as a1_mat * a2_mat, and that is scattered into a_dest. Every coupled-sector block is materialized (the matmul zeros the ones it does not reach), so the scatter overwrites a_dest in full.

A new matricizepermaliases predicate keeps the existing zero-copy path for a dense output already in matmul-aligned order, where matricizing a_dest returns a reshape/transpose view that mul! writes straight through. That case, and any nonzero β, still take the in-place mul! path. The predicate is pure (fusion style plus matricizekind) and defaults to false, so a graded output always takes the new path and a dense aligned output never does.

This trades one redundant write pass over the output for none, with allocations unchanged. On order-4 permuting graded contractions it measures around a 10% speedup, more when the matmul is a small fraction of the work and less as the matmul dominates.

[codecov]

Codecov Report

❌ Patch coverage is 90.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 83.76%. Comparing base (0890e18) to head (cf17b2e).

Files with missing lines	Patch %	Lines
src/matricize.jl	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #201      +/-   ##
==========================================
+ Coverage   83.74%   83.76%   +0.01%     
==========================================
  Files          24       24              
  Lines         732      739       +7     
==========================================
+ Hits          613      619       +6     
- Misses        119      120       +1     

Flag	Coverage Δ
docs	25.35% <55.55%> (+0.06%)	⬆️

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.