fix(ast): strip stack canary boilerplate from decompiled C output

[wizardengineer]

Add is_stack_canary_operation() to FunctionBuilder that detects compiler-inserted stack protector patterns by checking:

• Operations with inputs referencing __stack_chk_guard globals
• CALL operations targeting __stack_chk_fail

Detected operations are skipped during basic block creation, removing the canary entry load and fail call from the generated C.

This is a temporary solution that handles the straightforward cases (entry guard load and fail call) by pattern-matching on global names and call targets. The exit comparison block (if guard == saved) is not fully removed because the AST structuring engine reorganizes it into an if-else before our filter can catch the individual operations.

A complete solution would require value-level taint tracking — marking values derived from __stack_chk_guard and propagating that taint through copies, comparisons, and branches so the entire canary exit path can be identified and stripped. That approach is more invasive and should be discussed before implementing.

Partial fix for #161

[kyle-elliott-tob]

Two test/coverage gaps in the canary filter. Details inline.

[wizardengineer]

@kyle-elliott-tob — addressed both review threads and rebased onto main (the branch was conflicting against the create_basic_block → create_block_stmts rename from #237/#238).

• Test fixture (bl_device__process_entry.json): switched to CGEN-LABEL: bl_device__process_entry( + CGEN-NOT: __stack_chk_fail + CGEN-NOT: __stack_chk_guard. Label anchors to the function definition (not the forward decl) and the negation runs to EOF, so unordered_map emission order can't silently pass it.
• Switch-case inlining (OperationStmt.cpp): mirrored the canary skip in the inlined-case loop alongside the existing OP_BRANCH skip. The single_pred flag is currently false so the path is dead today, but the two emission paths now stay consistent if inlining is re-enabled.