[vm] Fix and test zero-copy BCS serialization (supersedes #20179, builds on #18535)

[gregnazario]

Description

Recreates the work from #20179 (which was accidentally closed after its branch was overwritten by unrelated commits). This carries #18535's zero-copy BCS serialization commits (serialize_ref / serialized_size_ref, resolving #14175) plus the fixups below. Because #18535 comes from a fork, its commits cannot be used as a PR base in this repo, so they are included here and this PR targets main.

Fixups on top of #18535:

1. Fix compile error (blocker). The new SerializationReadyValue<MoveTypeLayout, Container> impl passed the Arc<MoveStructLayout> from L::Struct directly where a &MoveStructLayout is expected, so move-vm-types did not compile. Added .as_ref() to match the existing Value serializer. (Not caught on [vm] Avoid deep copies in BCS serialization by using serialize_ref (#14175) #18535 because CI there is gated behind a failing permission-check, so all Rust build/test jobs were skipped.)

2. Restore the indexed-ref invariant guard. IndexedRef::read_ref enforces check_valid_for_indexed_ref, which rejects a stale/malformed indexed reference pointing at a nested container or another reference. The zero-copy path dropped this guard; it is restored so serialize_ref has the same safety behavior as the copy-based path.

3. Match read_ref depth accounting for indexed refs. IndexedRef::read_ref copies the referenced element at depth + 1, so the copy-based path enforces the nesting-depth limit one level deeper than the reference. The indexed-ref serialization arm now visits the element at self.depth + 1, so bcs::to_bytes / serialized_size reject at the same depth boundary as before (relevant for indexed refs to closures, whose captured arguments add nesting).

4. Add differential and depth-parity tests. Since the correctness requirement is that the reference path produces byte-identical output to the copy-based path (BCS output is consensus-sensitive), added tests asserting parity of bytes, size, and depth-limit behavior.

Disposition of automated review comments (from #20179)

• Depth-limit parity for indexed refs (closures): addressed by fixup (3) and pinned by serialize_ref_indexed_closure_enforces_depth_one_level_deeper.
• Failure-class change (read_ref errors now map to None → NFE_BCS_SERIALIZATION_FAILURE): intentional and consistent with the existing owned serialize path in [vm] Avoid deep copies in BCS serialization by using serialize_ref (#14175) #18535, which already maps every serializer error (including VM_MAX_VALUE_DEPTH_REACHED and invariant/tag failures) to None. Routing reference-structural failures through the same fail-closed abort keeps the reference and owned paths consistent; these are invariant violations that should not occur under a correct verifier. Left as-is by design.

How Has This Been Tested?

New tests in third_party/move/move-vm/types/src/values/serialization_tests.rs assert serialize_ref / serialized_size_ref match serialize / serialized_size across scalars, specialized/nested/general vectors, structs, nested structs, enum variants, signers (master/permissioned, legacy/non-legacy), delayed fields, indexed field/element refs, and depth-limit enforcement (incl. the indexed-ref-to-closure +1 parity).

cargo test -p move-vm-types --lib passes. cargo +nightly fmt and cargo clippy -p move-vm-types --lib --tests are clean for the changed code. aptos-move-stdlib (the natives caller) compiles.

Type of Change

• Bug fix
• Performance improvement
• Tests

Which Components or Systems Does This Change Impact?

• Move/Aptos Virtual Machine

Checklist

• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I tested both happy and unhappy path of the functionality