fix(LLVM): add --disable-non-volatile-memops and use it consistently

[marxin]

This PR introduces the --disable-non-volatile-memops CLI option, which preserves the use of the volatile keyword for memory operations (as it was prior to #6212). Additionally, it refactors how load and store operations are emitted, significantly simplifying and unifying the Operator::X expansion.

Fixes: #6401

[Copilot]

Pull request overview

This PR adds a CLI switch to control LLVM’s “non-volatile memory ops” behavior (restoring pre-#6212 trap-on-OOB semantics when disabled) and refactors LLVM codegen to centralize/standardize how loads, stores, and atomics are emitted and annotated.

Changes:

• Add --disable-non-volatile-memops (LLVM-only) and use it to conditionally enable enable_non_volatile_memops() in CLI engine configuration.
• Refactor LLVM translator load/store/atomic emission via new build_annotated_* helpers (and pass TargetData into the function code generator for size computations).
• Add a regression test for #6401 covering the “dropped SIMD load should still trap” behavior under volatile memops.

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated 3 comments.

File	Description
tests/compilers/issues.rs	Adds a regression test for #6401 validating trap behavior differences between volatile vs non-volatile memops in LLVM.
lib/compiler-llvm/src/translator/code.rs	Introduces helper builders for annotated load/store/atomics and refactors operator lowering to use them consistently.
lib/cli/src/backend.rs	Adds the --disable-non-volatile-memops runtime option and wires it into LLVM engine/config setup.