[copybara-service]

Shrink TransposeDecoder failure sentinel padding and use inline node storage

Reduces per-chunk setup overhead in riegeli::TransposeDecoder::Parse:

1. Shrinks the failure sentinel padding from 255 to
   kNumFailureSentinelNodes = 64 nodes (the maximum transition step offset is
   0xff >> 2 == 63, checked by a static_assert).
2. Replaces the heap-allocated std::vector<StateMachineNode> with inline
   storage in Context (inline_nodes[128]: up to 64 active nodes plus 64
   sentinels, ~3 KiB on Decode's stack), falling back to
   std::unique_ptr<StateMachineNode[]> for larger schemas, and deletes
   Context copy/move because state_machine_nodes may point into
   inline_nodes.
3. Explicitly initializes tag_data_size (and is_implicit on sentinels),
   which std::vector::resize previously value-initialized.
4. Passes absl::Span<const StateMachineNode> (active nodes only) to
   ContainsImplicitLoop and replaces its std::vector<size_t> with
   absl::FixedArray<uint32_t, Context::kInlineActiveNodes>.

[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.