C3: switch vxc to compile_pipeline; retire the AST middle/back-end after soak

Part of epic hiraditya/Vx.1#197. Roadmap:
docs/discussions/implementation_plans/flat_pipeline_convergence.md (C3). Depends on C2 (#200) at
parity.

Once C2 (#200) is at parity on the full suite:

• vxc drives src/pipeline.rs::compile_pipeline; codegen consumes the flat streams + frozen
  registry.
• The sequential AST middle/back-end moves behind --legacy-codegen for a soak.
• The parallel verification hooks (parallel_architecture_verifier) run in debug.
• After the soak, remove the AST middle/back-end.

This is the end state of the convergence: vxc production == the flat-array parallel pipeline, which
is the core architectural claim (docs/parallel_compiler_architecture.md). Fold in hiraditya/Vx.1#196 (assert
stream order determinism) before codegen depends on stream order.

Read docs/discussions/implementation_plans/flat_pipeline_convergence.md for C3, then inspect src/pipeline.rs::compile_pipeline and the architecture described in docs/parallel_compiler_architecture.md. Check C2 (#200) and the stream-order determinism work in hiraditya/Vx.1#196 before changing code. Done means vxc uses the flat pipeline, legacy codegen is available for the soak, debug verification hooks run, and the AST middle/back-end is removed afterward.