Intercept TO.tensoradd! instead of TensorKit.add_transform!

[lkdvos]

Problem

The blockwise tensoradd implementations here were only reachable through TensorKit internals:

• TO.tensoradd! delegated to permute!, which this package overrides, and
• the TK.add_transform! methods in src/tensors/indexmanipulations.jl caught the kernel.

Neither is a contract TensorKit owes us, and both stopped holding on TensorKit's
index-manipulation refactor (TensorKit.jl#526),
where TO.tensoradd! calls the braid kernel directly and add_transform! gained a conjsrc
argument (8 → 9 positional arguments, so all five of our methods silently became unreachable).

Nothing errors in that state. Dense block tensors survive via TensorKit's generic subblock
fallback, but sparse block tensors silently produce zeros, because the fallback writes through
blocks a sparse container never materialized:

W = Vtr[1] ⊗ Vtr[2] ⊗ Vtr[3] ← Vtr[4] ⊗ Vtr[5]
A = sprand(Float32, W, 0.5)
@tensor C[4, 5, 1, 3, 2] := A[1, 2, 3, 4, 5]   # norm(C) == 0

Measured against TensorKit#526, test/linalg/tensoroperations.jl fails, and TO.tensoradd! on a
SumSpace(ℂ^8, ℂ^8, ℂ^8) 4-leg block tensor also regresses 262 µs → 10473 µs.

Fix

Implement TO.tensoradd! for the block tensor types. That is the public entry point @tensor
lowers to, so it stays reachable however TensorKit arranges its kernels, and conjA is handled
explicitly rather than relying on it having been unwrapped into an adjoint beforehand.

The mixed block/plain methods take the concrete TensorMap so that
(BlockTensorMap, SparseBlockTensorMap) and the reverse resolve to the general method instead of
being ambiguous.

The five TK.add_transform! methods are left in place: they are still live against released
TensorKit's permute! routing, and our compat is TensorKit = "0.17". They can go when that
bound moves.

Tests

Two new testsets, both calling the entry points directly rather than through @tensor, since
the existing tests only exercised the macro and so pinned none of them:

• tensoradd! entry point — TO.tensoradd! over sparse/dense × two permutations × conjA.
• planar entry points — planaradd!, plus @planar trace and contraction. These are not
  currently broken (planaradd! → transpose!, planartrace! → trace_permute!,
  planarcontract! → contract! all still land on methods we override), but they rest on the same
  undocumented delegation, and there was previously no planar coverage here at all.

Verification

	TO.tensoradd!	@tensor permute	ambiguities	tests
TensorKit 0.17.1, before	260 µs	779 µs	0	pass
TensorKit#526, before	10273 µs	10712 µs	0	fail (zeros)
TensorKit 0.17.1, after	268 µs	769 µs	0	pass
TensorKit#526, after	269 µs	787 µs	0	pass

Aqua clean (including the method-ambiguity check) and runic clean.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 68.18182% with 7 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/tensors/tensoroperations.jl	68.18%	7 Missing ⚠️

Files with missing lines	Coverage Δ
src/tensors/tensoroperations.jl	85.22% <68.18%> (-5.69%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.