Validations against using EOF on pre-Osaka EVMs

[cameel]

I noticed that we're still missing validations against combinations like --experimental-eof-version 1 --evm-version cancun. I thought we had that safeguarded already because soltest does have this check, but apparently not. You'll either run into an ICE, or even worse - no error.

No error example

This finishes without errors:

echo 'contract C {}' | solc - --experimental-eof-version 1 --evm-version cancun

You only get an error if you request bytecode (Error: Experimental EOF support is only available for via-IR compilation.). And with --via-ir the compiler produces invalid IR.

ICE example

This triggers an ICE:

echo '{}' | solc --strict-assembly - --experimental-eof-version 1 --evm-version cancun

======= <stdin> (EVM) =======

Pretty printed source:
object "object" {
    code { { } }
}

Uncaught exception:
/solidity/libyul/backends/evm/EVMObjectCompiler.cpp(77): Throw in function void solidity::yul::EVMObjectCompiler::run(const solidity::yul::Object&, bool)
Dynamic exception type: boost::wrapexcept<solidity::yul::YulAssertion>
std::exception::what: Experimental EOF support is only available for optimized via-IR compilation and the most recent EVM version.
[solidity::util::tag_comment*] = Experimental EOF support is only available for optimized via-IR compilation and the most recent EVM version.

[cameel]

EVMHostPrinter seems to be only used by fuzzer so this wasn't breaking anything in practice (which is why it went unnoticed so far).

[cameel]

@aarlt Looks like EVM (and EOF) version is not accepted by assembly import on the CLI, even though it works in Standard JSON.