Read assertion from file with offsets

[Akuli]

The fix for ugly assertion reading. Adds file offsets to tokens, so that fseek() can be used to go to where the assertion starts and ends in the file. This way we can read the file contents after tokenizing.

• Fix weird error that only happens on Windows
• Port to self-hosted

[Akuli]

This doesn't work, because of the unread_byte() function and Windows CRLF (\r\n) line endings.

When tokenizing, it is possible to unread a byte that was previously read from the file. For example, to read the printf from printf("hi"), you read bytes one at a time until you get a byte that is not valid in a name, in this case (. Then you unread the ( so that the next read will return it.

The problem is unreading "\r\n" (Windows newline), which is translated to '\n' when reading, so that read_byte() consumes the \r and the \n, then returns just the \n. If we unread a \n, then we don't know whether the file contained \n or \r.

I could seek the file to check if it contains a \r before the \n, but then you can't compile unseekable files (such as output of another command), which would be annoying.

A better fix would be to read the entire source code into a string before tokenizing it.