[copybara-service]

Replace TransposeDecoder switch with preserve_none tail calls and frameless handlers

On 64-bit Clang x86-64 and AArch64 targets (RIEGELI_TAILCALL), replaces the
central HandleNode() switch dispatch in riegeli::TransposeDecoder::Decode
with 6-register preserve_none + musttail per-handler tail calls across 10
localized hunks (+412 / -1 lines in transpose_decoder.cc, with zero edits to
existing DecodingState callbacks or MaskBuffer overloads, and preserving the
compact 16B/24B StateMachineNode layout and untouched HandleNode() +
TransitionNode() switch loop on 32-bit and non-musttail targets):

1. Under #if RIEGELI_TAILCALL (defined(__clang__) && ABSL_HAVE_CPP_ATTRIBUTE(clang::musttail) && (defined(__aarch64__) || defined(__x86_64__)) && !defined(__ILP32__)), aligns StateMachineNode to
   32 bytes (alignas(32), zero dead padding) by adding a direct Callback
   function pointer alongside bool is_implicit : 1 and
   CallbackType callback_type (with Context's 128-node inline buffer,
   Context uses ~4.1 KB on Decode's stack). Other targets omit Callback
   and alignas(32), keeping StateMachineNode at 8 + 2 * sizeof(void*).
2. Dispatches via [[clang::preserve_none]] + [[clang::musttail]] handlers
   (falling back to the standard calling convention + [[clang::musttail]]
   under AddressSanitizer and HWAddressSanitizer and on non-Linux targets),
   pinning state, node, num_iters, transitions_cursor,
   transitions_limit, and dest in registers across transitions, pairing
   frameless fast-path wrappers
   (InvokeVarint, InvokeFixed, InvokeCopyTag, InvokeMessageStart,
   InvokeSubmessageStart, InvokeSubmessageEnd,
   InvokeSkippedSubmessageStart, InvokeSkippedSubmessageEnd) with
   ABSL_ATTRIBUTE_NOINLINE cold/slow helpers and InvokeSlowCallback, and
   #undefing all local RIEGELI_* macros at EOF.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.