Slow compilation (~3 s) of a 17-line contract on --via-ir and --via-ssa-cfg

## Description

A 17-line contract with deeply nested fixed-size arrays compiles in ~3 seconds with both `--via-ir --optimize` and `--experimental --via-ssa-cfg --optimize`. The legacy backend (without `--optimize`) compiles the same input in ~0.05 s, so the slowdown is roughly 60×. Found by AFL fuzzing of solc (the input was flagged as a hang against AFL's per-exec timeout).

Repro file also available as a gist: https://gist.github.com/msooseth/8bc721462a9f05ab754118d9579d62e4

## Environment

- Compiler version: `0.8.35-develop.2026.4.28+commit.b83005c9.Linux.g++` (commit `b83005c900d356e82f4d2da52be1601e1d8b3539`, tag `v0.8.35-pre.1-97-gb83005c90`)
- Compilation pipeline: `--via-ir` and `--via-ssa-cfg`
- Target EVM version: `prague`
- Framework/IDE: solc CLI
- EVM execution environment: n/a
- Operating system: Linux 6.19.13-arch1-1

Reproducer harness: `argotorg/solidity` pinned at `b83005c9` inside `solidity-fuzzing` @ `9aa104b07235022d9cc8e39bd5e3edb4f7d8df75`.

## Steps to Reproduce

```solidity
contract C {
struct S {
uint8[2][2][2][2][2][2][2][2] b;
}

S[][][] s1;
S[][1][] s2;
S[1][][1] s3;

function test1(S[][][] memory _a) public returns (S[][] memory){
s1.push();
s1[0] = _a[0];
return s1[0];
}
}
```

```bash
time solc --bin --via-ir --optimize --evm-version prague repro.sol # ~3.20 s
time solc --bin --experimental --via-ssa-cfg --optimize --evm-version prague repro.sol # ~3.00 s
time solc --bin --evm-version prague repro.sol # ~0.05 s
```

## Measurements

| Pipeline | Flags | Wall time |
|---|---|---|
| Legacy, no optimizer | `--bin --evm-version prague` | 0.05 s |
| via-IR + optimizer | `--bin --via-ir --optimize --evm-version prague` | 3.20 s |
| via-SSA-CFG + optimizer | `--bin --experimental --via-ssa-cfg --optimize --evm-version prague` | 3.00 s |

3 seconds on a 17-line input is enough that AFL classifies it as a hang.

Start by saving the reproduced contract as repro.sol and running the three solc CLI commands from the issue to confirm the timing gap. Trace the optimizer paths used by --via-ir and --via-ssa-cfg, then verify that the 17-line contract no longer causes a roughly three-second compile or AFL hang classification.

Written by the indexing model from the issue text.