Possibly unnecessary use of exponentiation in generated bytecode with legacy codegen

## Description

I have observed a use of exponentiation instruction that was a bit surprising. Here is a simple contract:

```solidity
contract A {
uint128[1] arr;

function set(uint128 v) external {
arr[0] = v;
}
}
```

Compiling this contract (`--bin-runtime`) with legacy codegen results in a bytecode with an exponentiation instruction, while compiling with `--via-ir` results in bytecode with no exponentiation.

This behaviour has been observed on versions `0.8.26` and `0.8.30`.

### Summary of matrix discussion:
The origin of the exponentiation of the instruction seems to be [here](https://github.com/ethereum/solidity/blob/406bdfb3abc8cfe73d59f9a748d53f9445591323/libsolidity/codegen/LValue.cpp#L315). I confirm that the constant base `0x100` is the value I have seen in the bytecode.
It seems this `EXP` is a substitute for a bit shift, and likely should have been updated for EVMs that support bit shifts.

I would add that exponentiation is difficult to handle by verification tools, so using bit shift would be very much appreciated :)

It would also be good to evaluate the impact of the change in bytecode size and gas consumption.
For this contract specifically, the legacy codegen produces smaller bytecode.

Start at libsolidity/codegen/LValue.cpp around line 315, where the exponentiation originates, and compile the example with --bin-runtime using legacy codegen and --via-ir for comparison. Determine whether replacing the constant-base EXP with a bit shift changes bytecode size or gas consumption, and verify the resulting bytecode for the example.

Written by the indexing model from the issue text.