Motivation

• Separate the parsing of CoreDSL instruction metadata from LLVM IR emission so the front-end only records behavior structure while the back-end is responsible for generating IR, preserving existing semantics but enabling a clear frontend/backend split.

Description

• Added behaviorTokens and behaviorLine to CDSLInstr and included Token.hpp so the front-end can record the token stream and starting line for each behavior: block (file: InstrInfo.hpp).
• Introduced ParseCoreDSL2FrontEnd that parses instruction metadata and records behavior tokens, and GenerateBehaviorIR that replays those tokens to generate LLVM IR; updated public API in Parser.hpp and replaced the single-stage call in Main.cpp with the two-stage pipeline.
• Reworked ParseBehaviour to stop recording tokens instead of emitting IR and implemented GenerateInstructionBehaviorIR which constructs a TokenStream in replay mode to re-run the behavior parsing/IR emission code (file: Parser.cpp).
• Extended TokenStream with a replay mode and a constructor TokenStream(std::string&&, std::vector<Token>&&, int) so the backend can parse from the recorded token vector while preserving identifier indices and line numbering (files: TokenStream.hpp and TokenStream.cpp).
• Added a small validation in the behavior IR generator to detect leftover tokens after parsing and surface a syntax error if present.

Testing

• Ran git diff --check on the changes to ensure there are no whitespace/errors reported by the diff tooling and it completed with no issues.

Codex Task