[executorch][native] Add deserializer bridge (Program -> Method)

[SS-JIA]

Stack from ghstack (oldest at bottom):

• -> [executorch][native] Add deserializer bridge (Program -> Method) #22105
• [executorch][native] Add Method + ValueRole to the in-memory IR #22104
• [executorch][native] Add Graph index arena to the in-memory IR #22103
• [executorch][native] Add Node to the in-memory IR #22102
• [executorch][native] Add Argument tagged union to the in-memory IR #22101
• [executorch][native] Add Value + index-arena ids to the in-memory IR #22100
• [executorch][native] Add Scalar value type to the in-memory IR #22099
• [executorch][native] Add TensorMeta + ScalarType in-memory IR types #22098
• [executorch][native] Add native runtime Program reader + Graphviz visualizer #22097
• [executorch][native] Add native_executor test harness #22096

Threads the in-memory IR through the program reader: ptn::Program now
materializes a serialized native_backend::Program into ptn::Method /
Graph / Node / Argument / Value. This is the bridge from the FlatBuffer
wire form to the in-memory graph the rest of the runtime will consume.

Loading is lazy and keyed by name, mirroring ExecuTorch's own
Program::load_method(name) model:

• Program::load() parses and verifies the buffer only; it no longer eagerly
  builds anything.
• Program::num_methods() / method_names() read the FlatBuffer directly.
• Program::get_method(name) materializes the method's IR on first request and
  caches it (a mutable unordered_map<string, Method>, so lookups work on a
  const Program and returned references stay stable across later insertions);
  subsequent calls return the same instance. Throws if the name is absent.

The deserializer itself lives in a new Deserialize.cpp, keeping Program.cpp
the thin reader. Per method it:

• Builds the Graph: creates a Value per SSA name (Tensor when the tensor
  side table carries metadata, else a None value for scalar / symbolic outputs),
  a Node per fx node with its Arguments (resolving in-graph SSA names to
  arena ValueRefs: tensor refs, symbolic scalar refs, tensor / optional-tensor
  lists, getitem-folded Tensor[] outputs into elem_refs), the graph I/O
  ref lists, then reset_schedule() + rebuild_def_use(). Recurses HOP
  subgraphs (GraphArg) into the per-graph subgraph arena.
• Builds the method bindings: constants and mutable_buffers become
  DataBindings (stamping each placeholder Value's role / data_key;
  graph inputs not otherwise bound become UserInput), and output_specs
  resolve their string target to a target_ref (an SSA name for
  UserInputMutation, an fqn for BufferMutation).

Also adds a --methods mode to native_executor that dumps each materialized
method via Method::to_string().

Pure std + flatbuffers only (no ExecuTorch dependency).

Differential Revision: D114426391

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22105

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 1 Unrelated Failure

As of commit 3194b01 with merge base 9a2d135 ():

NEW FAILURE - The following job has failed:

• pull / test-custom-ops-linux / linux-job (gh)
  ##[error]fatal: unable to access 'https://github.com/gflags/gflags.git/': Error in the HTTP2 framing layer

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• pull / test-qnn-testsuite-linux / test-backend-linux (qnn, models) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.