[GR-65450] Port ML-KEM stubs

[graalvmbot]

Problem

ML-KEM intrinsics currently depend on HotSpot-owned stubs. Graal needs its own lowering and stub linkage so the Kyber operations used by ML-KEM remain intrinsified under Graal JIT and Native Image, with emitted code kept in sync with the HotSpot implementations on AMD64 and AArch64.

Summary

This PR moves ML-KEM stub ownership into Graal while keeping the emitted bodies structurally aligned with the HotSpot stubs.

• Adds Kyber graph nodes, plugin wiring, foreign-call linkage, and LIR generator support for the ML-KEM stub operations.
• Ports the AMD64 and AArch64 Kyber stubs for NTT, inverse NTT, NTT multiplication, add-poly, 12-to-16 parsing, and Barrett reduction.
• Adds the required assembler support and architecture-specific LIR ops for the new stubs.
• Registers the stubs for Native Image and keeps CPU-feature gating aligned with the HotSpot paths, including ASIMD gating on AArch64.
• Preserves HotSpot comments and @SyncPort provenance in the ported support code.