[CubeCL][P1.2] Implement broadcasting, elementwise operators, and dtype conversion

Objective

Implement existing unary/binary elementwise operations, broadcasting, comparison/logical behavior and supported dtype conversion over the CubeCL foundation.

Acceptance criteria

• A checked operation inventory maps every current public operation to implemented, unsupported-with-reason, or separately owned status.
• Host/CubeCL property fixtures cover broadcasting, scalar operands, mixed layouts, zero dimensions, NaN/infinity, signed/unsigned boundaries and conversion failures.
• Numerical comparison uses operation/dtype-specific tolerances rather than one global threshold.
• Unsupported dtypes or runtime capabilities fail visibly.
• Cold compilation and warm execution are measured separately for representative chains.
• No deferred fusion behavior is introduced in this issue.

Parent: #35

Tembo execution contract

automation:
  eligible: true
  executor_profile: tembo
  readiness: blocked
  primary_repository: TinyChain-Inc/ha-ndarray
  base_branch: main
  writable_scope:
    - src/cubecl/elementwise/**
    - src/cubecl/cast/**
    - src/ops.rs
    - tests/conformance/**
    - benches/**
  prerequisites:
    - "#50 designated accelerator validation lane"
    - "#41"
  concurrency_group: cubecl-elementwise
  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

Begin by checking prerequisites #50 and #41, then inspect src/cubecl/elementwise/, src/cubecl/cast/, src/ops.rs, tests/conformance/, and benches/. Run the required Cargo format and test commands before mapping the operation inventory. Done means the listed conformance, failure-handling, compilation, and benchmark evidence is complete without deferred fusion.