Add switchable optimization config and benchmark framework

WaferConfig: unified config controlling all optimizations individually.
ForthVM::new_with_config(config) to create VMs with custom optimization settings.
All 8 switchable optimizations: peephole, constant_fold, strength_reduce, dce,
tail_call, inline (IR passes) + stack_to_local_promotion (codegen).

Benchmark framework (crates/core/tests/benchmark_report.rs):
- 7 Forth benchmarks: Fibonacci, Factorial, SumRecurse, NestedLoops, GCD, MemFill, Collatz
- Correctness verification across all configs (runs in CI)
- Full report with 128 optimization combinations (cargo test --ignored)
- Measures execution time, compilation time, WASM module bytes
- CONSOLIDATE impact comparison

Key findings from benchmark report:
- Inlining: -77% exec time on Fibonacci, -92% on Collatz
- Stack-to-local promotion: -5.5% WASM module size
- CONSOLIDATE: -72% exec time on Fibonacci (call_indirect -> direct call)
- All optimizations combined: best overall performance

crates/core/src/codegen.rs

@@ -85,6 +85,8 @@ pub struct CodegenConfig {
     pub base_fn_index: u32,
     /// Number of functions already in the table.
     pub table_size: u32,
+    /// Enable stack-to-local promotion for straight-line words.
+    pub stack_to_local_promotion: bool,
 }
 
 /// Result of compiling a word to WASM.
@@ -1457,7 +1459,7 @@ pub fn compile_word(
 
     // -- Code section --
     // Determine whether to use stack-to-local promotion
-    let promoted = is_promotable(body);
+    let promoted = config.stack_to_local_promotion && is_promotable(body);
     let scratch_count = count_scratch_locals(body);
     let num_locals = if promoted {
         let (preload, _) = compute_stack_needs(body);
@@ -1900,6 +1902,7 @@ mod tests {
         CodegenConfig {
             base_fn_index: 0,
             table_size: 16,
+            stack_to_local_promotion: true,
         }
     }
 
@@ -2123,6 +2126,7 @@ mod tests {
         let cfg = CodegenConfig {
             base_fn_index: 7,
             table_size: 16,
+            stack_to_local_promotion: true,
         };
         let m = compile_word("t", &[IrOp::PushI32(1)], &cfg).unwrap();
         assert_eq!(m.fn_index, 7);