at-dispatch-v2

AT_DISPATCH_V2 conversion transforms legacy AT_DISPATCH_ALL_TYPES_AND* macros to v2 format by reordering arguments, wrapping lambdas with AT_WRAP, expanding type groups with AT_EXPAND, and adding #include <ATen/Dispatch_v2.h> to aten/src/ATen/native/ kernel implementations.

AT_DISPATCH_V2 provides safer, more scalable kernel dispatch with explicit type enumeration and structured argument ordering, replacing the older AT_DISPATCH macro patterns that relied on type group macros like AT_DISPATCH_ALL_TYPES_AND*.

AT_DISPATCH_V2 migration applies specifically to AT_DISPATCH_ALL_TYPES_AND*, AT_DISPATCH_FLOATING_TYPES*, and related dispatch macros in ATen kernel operator implementations; legacy patterns should be converted to maintain consistency with PyTorch's dispatch infrastructure.

Type groups expand using AT_EXPAND during AT_DISPATCH_V2 conversion, replacing macro-based type grouping with explicit individual type listings such as kBFloat16, kHalf, and kBool for precise kernel dispatch.

AT_DISPATCH_V2 integration requires converting macro invocations in ATen kernel operator implementations, wrapping dispatch logic with AT_WRAP, and ensuring the Dispatch_v2.h header is included for compatibility with modern PyTorch dispatch patterns.

Conversions target aten/src/ATen/native/ directory files containing AT_DISPATCH macro invocations in kernel operator implementations, updating macro signatures and adding required dispatch v2 headers.