[executorch][native] Add Method + ValueRole to the in-memory IR

[SS-JIA]

Stack from ghstack (oldest at bottom):

• -> [executorch][native] Add Method + ValueRole to the in-memory IR #22281
• [executorch][native] Add Graph index arena to the in-memory IR #22280
• [executorch][native] Add Node to the in-memory IR #22279
• [executorch][native] Add Argument tagged union to the in-memory IR #22278
• [executorch][native] Add Value + index-arena ids to the in-memory IR #22277
• [executorch][native] Add Scalar value type to the in-memory IR #22276
• [executorch][native] Add TensorMeta + ScalarType in-memory IR types #22275
• [executorch][native] Add native runtime Program reader + DOT visualizer #22274

Adds ptn::Method, the stateful wrapper around one top-level Graph, and the
per-value ValueRole classification it drives. Graph is a pure function body;
Method binds its placeholder and output values to the things outside the pure
dataflow — external constant data, non-persistent buffers, and the per-output
classification that separates a real result from a state writeback. It mirrors
the schema Method but resolves every in-graph reference to a ValueId.

Method lives at runtime/ (peer to the Program reader), above the graph/
arena package, since it is a method-level rather than a graph-level concept.

New types:

• ValueRole (on Value, in graph/Value.h): UserInput / Parameter / Buffer /
  ConstantTensor / Intermediate. The placeholder roles are pinned to the schema
  InputKind ids; Intermediate extends them for a value a node produces.
  Stamped by the deserializer, it gives an O(1) per-value answer to "how is this
  value's storage owned / sourced" (the memory planner and runtime binder are
  the eventual consumers).

• DataBinding (schema NamedTensorRef + MutableBufferSpec, merged): binds a
  placeholder ValueId to external storage. key is the fqn (an
  external-constant-file key when has_data, else the buffer's cross-method
  identity), and the binding is the only place it lives; has_data selects the
  load path (fetch bytes vs zero-init at load); mutated marks a Buffer written
  in place whose state persists across executions; role mirrors the bound
  value's role for direct access when iterating bindings.

• OutputKind (graph-level: UserOutput / BufferMutation / UserInputMutation) and
  OutputSpec (aligned to graph.output_ids by index). A mutation output's
  target_id is a ValueId to the mutated placeholder value in this graph
  (both the mutated user input and the mutated buffer are lifted placeholders
  here), replacing the schema's dual-namespace target string; the wire fqn /
  SSA name is recovered from that value.

• Method itself: name, Graph graph, data_bindings, output_specs. HOP
  subgraphs carry no bindings (their params are lifted here and passed as
  operands), which is why Method wraps Graph rather than folding into it.

• to_string(const Method&) (utils/Print.{h,cpp}, target runtime:print) —
  the method layer of the IR printer, split from the graph-level one only
  because Method sits above the graph/ package.

Pure std only (no ExecuTorch, no flatbuffers).

Differential Revision: D114426388

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22281

• 📄 Preview Python docs built from this PR

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

✅ You can merge normally! (1 Unrelated Failure)

As of commit 809a1f7 with merge base a5f15b5 ():

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.

[digantdesai]

Review automatically exported from Phabricator review in Meta.