better handling of empty array from inferLanguage

[Janther]

There are 2 cases in which we receive an empty array from inferLanguage.

• the range is outside of the supported versions.
• there are ranges in the pragmas that don't intersect.

with this in mind we are defaulting to the latest supported version and return a more detailed error.

also in this PR:

I could not parse any nightly builds even when using specific versions of nightly builds of solc. Therefore I'm removing this test.

NomicFoundation/slang#1346

[fvictorio]

I'm not sure I prefer the new behavior. I'd personally prefer to always throw an error:

• If you are doing pragma solidity ^0.7.0; pragma solidity ^0.8.0; then this is an error. Parsing it anyway, even if it works, doesn't seem ideal to me (even if that's what the antlr-based parser did).
• If you are doing pragma solidity 0.8.99 and that version is not supported by Slang, I think I'd also prefer to error.

The very first improvement I would do here is to try to tell apart between those two scenarios, so that users can know what's going on. I know that's not trivial though.

Then if people complain that this is too restrictive, we can consider fallback versions like you are doing here.

I don't feel super strongly about this though. If you do, and think this is the best approach, then feel free to merge.

[Janther]

the issue I'm trying to fix here is when there is a new solidity version and slang takes some time in releasing the support and then we take some time on adapting to this, and finally the user installs our latest release.
With the current solution we won't even attempt to format even if the newest solidity version has no syntax changes or the developer is not using the newest syntax.