Duplicated error definitions in the ABI

## Description

If the same error is defined in different files within the inheritance/use chain, it ends up twice in the ABI despite having the same selector.
The generated bytecode seems fine as I couldn't find the selector being duplicated in it.

## Environment

- Compiler version: 0.8.20
- Target EVM version (as per compiler settings): London
- Framework/IDE (e.g. Truffle or Remix): Foundry
- EVM execution environment / backend / blockchain client: N/A
- Operating system: MacOS

## Steps to Reproduce

```solidity
pragma solidity 0.8.20;

library A {
error CustomError();

function iCustomError() internal pure {
revert CustomError();
}
}

library B {
error CustomError();

function iCustomError() internal pure {
revert CustomError();
}
}

contract SolidityBug {
function foo() public pure {
A.iCustomError();
}

function bar() public pure {
B.iCustomError();
}
}
```

will generate

```json
"abi": [
{
"inputs": [],
"name": "CustomError",
"type": "error"
},
{
"inputs": [],
"name": "CustomError",
"type": "error"
},
{
"inputs": [],
"name": "bar",
"outputs": [],
"stateMutability": "pure",
"type": "function"
},
{
"inputs": [],
"name": "foo",
"outputs": [],
"stateMutability": "pure",
"type": "function"
}
]
```

I know this seems silly, but the real world example on how this can happen is the following. Here the `AddressEmptyCode` error will be defined in both OZ libs, and it'll end up twice in the ABI. (need to use the latest master from OZ)

```solidity
import "@openzeppelin/contracts/token/ERC20/utils/SafeERC20.sol";
import "@openzeppelin/contracts-upgradeable/proxy/utils/UUPSUpgradeable.sol";

contract SolidityBug is UUPSUpgradeable
{
function _authorizeUpgrade(address newImplementation) internal override {}

function useSafeERC20(IERC20 token, address to, uint256 amount) external {
SafeERC20.safeTransferFrom(token, msg.sender, to, amount);
}
}
```

Start by reproducing the Solidity 0.8.20 example and tracing ABI generation for error declarations from the two libraries. Identify the compiler entry point and existing ABI tests involved, then add a regression test showing that identical error selectors appear only once in the generated ABI.

Written by the indexing model from the issue text.