One host runtime, step 2: cuda-oxide adopts the root cuda-core types and cuda_core::simt goes away #1460

Remove the duplicate cuda_core::simt host runtime so Tile and SIMT code share allocation, context, stream, launch, and resource types. This builds on #1459 and coordinates with the async migration in #1461.

Requirements

• Share the allocation implementation and retain DeviceBuffer<T> as the typed wrapper. Preserve owned, borrowed, foreign, and sliced storage semantics; coordinate borrowed subviews with feat(cuda-core): add borrowed DeviceBuffer subviews #1447.
• Keep DType and DeviceCopy independent. Use DeviceCopy for general raw-buffer operations, DType for tensor/scalar APIs, and DType + DeviceCopy only where both contracts are needed.
• Add the missing compatible scalar-wrapper DeviceCopy implementations: tf32, f8e4m3fn, f8e5m2, f8e8m0fnu, f8e5m3fnu, and f4e2m1fnx2, with safety justifications.
• Implement DeviceCopy for bool and keep bool: DType. bool does not implement bytemuck::AnyBitPattern, and the safe reinterpreting casts (DeviceBuffer::cast_elem, DeviceBuffer::cast_chunks, Tensor::reinterpret) are bounded on AnyBitPattern, so a bool buffer can only come from memory written as bool. This is Safe element casts can produce invalid bool values (Tensor::reinterpret, DeviceBuffer::cast_elem/cast_chunks) #1506 and lands before the rest of this issue. Storage and padding/packing policy for packed and sub-byte types are tracked in cuda-core: define boolean storage and element padding/packing policy #1503.
• Migrate CudaStream to shared Stream and unify the context API while keeping device and context identities distinct. Resolve the type model and any convenience pairing in cuda-core: distinguish CUDA devices and contexts in the shared host API #1504; coordinate global-context handling with Make with_global_device_context state explicit (Uninitialized | Available | Borrowed) #1403.
• Do not implicitly synchronize GPU work on drop. Preserve resource lifetimes until submitted work completes, including early-dropped async work and borrowed views. Coordinate cleanup and lifetime changes with One host runtime, step 3: cuda-oxide adopts the root cuda-async types and cuda_async::simt goes away #1461.

Implementation

• Integrate shared typed storage and per-operation trait bounds, including PinnedHostBuffer.
• Migrate context, stream, event, module, and function users to the shared API defined by cuda-core: distinguish CUDA devices and contexts in the shared host API #1504.
• Share LaunchConfig and launch_kernel* helpers while preserving launch modes, validation, and resource checks.
• Specify the shared launch-query API signatures and implement them in independently reviewable changes.
• Reconcile simt::vmm, peer, memory, and embedded with shared modules.
• Implement completion-safe cleanup without synchronization on drop.
• Migrate cuda-oxide crates, examples, generated-project templates, and documentation. Retire or replace check-host-api-paths.sh when its SIMT-specific spelling is obsolete.
• Move simt_* tests into shared suites and delete cuda-core/src/simt/ after consumers are migrated.

Acceptance

Both programming models use the shared types, and the temporary SIMT module is removed. Validate allocation identity and zero-copy sharing between a Tile tensor and a SIMT launch argument, including view offsets, context compatibility, ordering, and completion lifetimes. Cover positive and negative trait bounds and boolean validity. Document breaking API changes and their migration path.

Related: #1461 (async runtime), #1503 (element representation), #1504 (device/context model), #1505 (public namespaces), #1506 (bool and the safe casts).