[CubeCL][P2.2] Implement matrix multiplication and current linear-algebra operations

Objective

Implement matrix multiplication and the current public linear-algebra operations on CubeCL without assuming NVIDIA-specific topology.

Acceptance criteria

• Inventory maps every current public linalg operation and supported shape/dtype combination.
• Matmul covers vector/matrix/batched forms actually supported, non-contiguous inputs, degenerate shapes and dimension failures.
• Kernels derive topology from CubeCL runtime capabilities rather than hard-coded warp dimensions.
• Results match host references within documented operation/dtype tolerances.
• Cold compilation, warm throughput, temporary memory and correctness are reported on CPU and a designated accelerator.
• Any use of CubeCL library kernels versus project kernels is documented with compatibility implications.

Parent: #36

Tembo execution contract

automation:
  eligible: true
  executor_profile: tembo
  readiness: blocked
  primary_repository: TinyChain-Inc/ha-ndarray
  base_branch: main
  writable_scope:
    - src/cubecl/linalg/**
    - src/ops.rs
    - tests/conformance/**
    - benches/**
  prerequisites:
    - "#50 designated accelerator validation lane"
    - "#41"
    - "#42"
  concurrency_group: cubecl-linalg
  required_commands:
    - "cargo fmt --all -- --check"
    - "cargo test --no-default-features --features cubecl"
    - "cargo test --all-features"
  evidence:
    cpu_conformance: required
    accelerator_conformance: required-on-designated-runner
    actual_browser: not-required
    benchmark_report: required
    exact_hardware_and_driver_identity: required
  external_actions: none
  draft_pr_only: true
  completion_authority: human
  budget:
    size: large
    max_revisions: 3
    max_wall_minutes: 180

Start by reviewing src/cubecl/linalg/** and src/ops.rs, then inspect tests/conformance/** and benches/** to identify the current public operations and existing coverage. Run the required cargo fmt and test commands; completion requires CPU and designated-accelerator conformance, benchmark evidence, hardware and driver identity, and documented kernel compatibility.