feat(runtime): implement F64 instructions and fix CallIndirect LEB128 parsing

- Add comprehensive F64 instruction support in stackless engine:
  - Memory operations: F64Const, F64Load, F64Store
  - Arithmetic: F64Add, F64Sub, F64Mul, F64Div
  - Comparisons: F64Eq, F64Ne, F64Lt, F64Gt, F64Le, F64Ge
  - Unary: F64Abs, F64Neg, F64Ceil, F64Floor, F64Trunc, F64Nearest, F64Sqrt
  - Binary: F64Min, F64Max, F64Copysign
  - Conversions: F64ConvertI32S/U, F64ConvertI64S/U, F64PromoteF32
  - Reinterpret: F64ReinterpretI64, I64ReinterpretF64

- Fix CallIndirect instruction parsing to properly read table_idx as LEB128
  instead of assuming single byte (was causing instruction stream corruption)

- Disable bump allocator in favor of cabi_realloc for proper dlmalloc headers
  (prevents memory corruption when Rust frees allocated strings)

These fixes enable calculator.wasm to correctly compute: 5+3=8, 10-7=3, 6*7=42, 100/4=25

Author: avrabe

wrt-runtime/src/instruction_parser.rs

@@ -236,11 +236,13 @@ fn parse_instruction_with_provider(
             Instruction::Call(func_idx)
         },
         0x11 => {
-            // CallIndirect
-            let (type_idx, bytes) = read_leb128_u32(bytecode, offset + 1)?;
-            consumed += bytes;
-            consumed += 1; // Skip table index (always 0 in MVP)
-            Instruction::CallIndirect(type_idx, 0)
+            // CallIndirect: type_idx (LEB128 u32) followed by table_idx (LEB128 u32)
+            // Note: table_idx can be multi-byte LEB128, not always single byte!
+            let (type_idx, type_bytes) = read_leb128_u32(bytecode, offset + 1)?;
+            consumed += type_bytes;
+            let (table_idx, table_bytes) = read_leb128_u32(bytecode, offset + 1 + type_bytes)?;
+            consumed += table_bytes;
+            Instruction::CallIndirect(type_idx, table_idx)
         },
 
         // Parametric instructions