[maleadt]

Metal.jl gets sparse vectors and matrices. GPUArrays 13 implements them for every back-end (JuliaGPU/GPUArrays.jl#745), and this PR names them for Metal storage and moves host sparse arrays to the GPU.

using Metal, SparseArrays, LinearAlgebra

A = mtl(sprand(1000, 1000, 0.01))  # MtlSparseMatrixCSR{Float32, Int32}
x = Metal.rand(Float32, 1000)

A * x            # sparse × dense
2 .* A .+ A      # broadcasting stays sparse
A * A'           # sparse × sparse
A[1:10, :]       # indexing
SparseMatrixCSC(A)  # back to the host

• MtlSparseVector and MtlSparseMatrixCSR/CSC/COO are aliases of the GPUArrays types with MtlVector storage.
• mtl(::SparseMatrixCSC) gives CSR, the best format for A * x. It uses Int32 indices when they fit, and narrows Float64 values like it does for dense arrays. adapt(MtlArray, A) keeps the format and types.
• The constructors on the aliases convert host input on the host first, so Float64 data works. MtlArray(A) densifies on the device.
• A new docs page covers the transfer policy and which format suits which operation.
• Metal's broadcast specialization now only applies to GPU destinations, which avoids an ambiguity with in-place broadcasts into sparse arrays.

There is no vendor sparse library, so all products use GPUArrays' portable kernels. This is a minor release (1.13.0); Metal had no sparse support before. It supersedes #909.

Tested on an M1 and an M3 Pro (Julia 1.13): Metal's sparse tests (53) and GPUArrays' shared sparse testsuite on MtlArray (12,365), against GPUArrays tb/sparse. The package also loads on Julia 1.10.

The last commit (TEMP) points CI at the unregistered GPUArrays branch. Drop it before merging.

[codecov]

Codecov Report

❌ Patch coverage is 86.95652% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.68%. Comparing base (806590b) to head (492d83b).

Files with missing lines	Patch %	Lines
src/sparse.jl	85.71%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #982      +/-   ##
==========================================
- Coverage   88.86%   83.68%   -5.18%     
==========================================
  Files         105      105              
  Lines        9399     8843     -556     
==========================================
- Hits         8352     7400     -952     
- Misses       1047     1443     +396     

☔ 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.