[FIRRTL] SizeOf intrinsic requires optimizations enabled to successfully lower

```firrtl
FIRRTL version 4.0.0
circuit Top:
public module Top:
output size: UInt<32>
wire empty: {nested: {zero: UInt<1>[0]}}
node sizeNode = intrinsic(circt_sizeof : UInt<32>, empty)
connect size, sizeNode
```

With normal optimization enabled, this succeeds because canonicalization folds the intrinsic to `UInt<32>(0)`:

```text
module Top(
output [31:0] size
);

assign size = 32'h0;
endmodule
```

Running `firtool input.fir --disable-opt ` where optimization is disabled, the intrinsic remains in the IR and `LowerToHW`
reports:

```text
test.fir:6:21: error: SizeOf should have been resolved.
node sizeNode = intrinsic(circt_sizeof : UInt<32>, empty)
^
test.fir:6:21: note: see current operation: %6 = "firrtl.int.sizeof"(%5) : (!firrtl.bundle, 0>>>) -> !firrtl.uint<32>
test.fir:6:21: error: 'firrtl.int.sizeof' op LowerToHW couldn't handle this operation
node sizeNode = intrinsic(circt_sizeof : UInt<32>, empty)
^
test.fir:6:21: note: see current operation: %6 = "firrtl.int.sizeof"(%5) : (!firrtl.bundle, 0>>>) -> !firrtl.uint<32>
```

`SizeOfIntrinsicOp` is a compile-time-only operation. Its folder computes the bit width from the operand type; it has no hardware lowering. Therefore, `LowerToHW` requires the canonicalizer to have resolved it first. We should be able to successfully compile this without optimizations.

Reproduce the issue with the FIRRTL input and `firtool input.fir --disable-opt`. Start at `SizeOfIntrinsicOp` and its folder, then inspect the `LowerToHW` entry point to determine how unresolved SizeOf operations are handled. Done means the example compiles successfully with optimizations disabled and produces the expected zero-valued output.

Written by the indexing model from the issue text.