Shed registry-blocking deps, add per-position fusion styles, switch to explicit factorization names

[mtfishman]

Summary

Bumps TensorAlgebra to 0.10.0, a breaking release that prepares it for the General registry and reworks two pieces of the interface.

Drops the FunctionImplementations and TypeParameterAccessors dependencies, the only two things keeping TensorAlgebra out of the General registry. TensorAlgebra now owns the shared in-place extension points zero! and scale! and the lazy permuteddims hook, declared public so downstream array types can overload them directly. The two small TypeParameterAccessors helpers used internally are inlined as thin Base wrappers. Also removes the over-broad getindex(::AbstractArray, ::AbstractBlockPermutation) shorthand, which was type piracy and an ambiguity trap with no external users, in favor of the existing blockpermute.

Gives the Matricize contraction algorithm three fusion styles (left_fusion_style, right_fusion_style, output_fusion_style), one for each reshape position, instead of a single shared one. A single-style constructor copies to all three, so the symmetric common case and existing callers are unchanged. This lets a backend that needs one position to behave differently (the fermionic contraction twist on the second operand in GradedArrays) pick a distinct style at algorithm selection without re-overloading the whole routine.

Switches the tensor-level factorizations from the catch-all names (svd, qr, lq, eigen, eigvals, svdvals, polar, orth, factorize) to the explicit names MatrixAlgebraKit uses (svd_compact, svd_full, svd_trunc, qr_compact, qr_full, eigh_full, eig_full, and the rest), so the variant lives in the name rather than behind a full, trunc, or ishermitian keyword. These are TensorAlgebra-owned functions taking the codomain/domain partition signature, out-of-place only. The MatrixAlgebra submodule sheds the matrix-level wrappers that re-exposed those catch-alls and keeps only what MatrixAlgebraKit does not provide (gram_eigh_full, the safe-clamping helpers, the degenerate-truncation strategy).

[codecov]

Codecov Report

❌ Patch coverage is 97.91667% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 81.56%. Comparing base (2783b56) to head (fa29830).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/inplace.jl	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #180      +/-   ##
==========================================
+ Coverage   81.37%   81.56%   +0.18%     
==========================================
  Files          26       27       +1     
  Lines        1047      987      -60     
==========================================
- Hits          852      805      -47     
+ Misses        195      182      -13     

Flag	Coverage Δ
docs	28.06% <27.08%> (+1.86%)	⬆️

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.