Make lowering failure tests less fragile

[nathanwhit]

The current lowering failure tests check that the error message resulting from the lowering attempt matches a given string exactly. This is fine in most cases, but tests which expect parsing errors are incredibly fragile in that the contents of their error messages change whenever the lexer or parser is modified.

This PR makes a small change so that lowering failure tests which produce a parsing error only have to provide the expected error kind (such as "UnrecognizedToken", etc.) instead of a full error message, which makes the tests less fragile to parser/lexer changes.

It's not exactly a beautiful fix, but it at least alleviates the immediate annoyance of tests frequently breaking.

[AzureMarker]

The test macro allows you to only give the prefix of the output you expect (ex. "Unique") by using assert_same:

chalk/tests/test/mod.rs

Line 35 in 26242c0

fn assert_same(result: &str, expected: &str) {

We could do the same thing for lowering tests, and then we wouldn't need the special parse_error syntax and it would work for non-parse errors (assuming they exist). It should also be backwards-compatible with the existing tests.

lowering_error! {
    // ...
    error_msg {
        "parse error: UnrecognizedToken"
    }
}

[jackh726]

I agree with @Mcat12 here: we should just check whether the message starts with the expected text. That assert_same function can just be moved into test_util

[nathanwhit]

Sounds reasonable to me. I've updated the PR to make that change.