[5.9] Fix 5 parser bugs

[ahoppen]

• Explanation: Add a new test mutation strategy to flip the presence of token in the assertParse tests (fixes 5 parser bugs) #1706 uncovered 5 parser bugs with different severities. Fix them and add the testing strategy that found them to release/5.9

1. A round-trip failure for unexpected modifiers in a tuple type

func test(a: (borrowing F o))

2. A round-trip failure for modifiers in front of an editor placeholder in declaration position

struct a {
  public<#declaration#>
}

3. A precondition failure when the primary associated type clause of a protocol is empty

protocol X<> {}

4. It also uncovered an issue where we weren’t able to retrieve the trivia pieces of a token after it had been modified using with because after the modification the token was a parsedToken that no longer resided in a ParsingSyntaxArena. The fix is to search through an arena’s childRefs to find the ParsingSyntaxArena that created it. @rintaro can you review this change?
5. It uncovered that we were making assumptions about the presence of tokens in ParseDiagnosticGenerator, mostly that tokens in UnexpectedNodesSyntax are present. While this assumption is valid for trees generated by SwiftParser, it doesn’t need to hold for manually generated trees. A couple of minor adjustments to filter for only present tokens fixes this.

• Scope: Each fix is pretty targeted
• Risk: The previous behavior was plainly wrong/crashed. So I think it’s very acceptable.
• Testing: Fixed all issues found by the new testing strategy
• Issue: rdar://109783066
• Reviewer: @rintaro and @bnbarham on Add a new test mutation strategy to flip the presence of token in the assertParse tests (fixes 5 parser bugs) #1706

[ahoppen]

@swift-ci Please test

[ahoppen]

@swift-ci Please test

[ahoppen]

@swift-ci Please test Windows

[ahoppen]

@swift-ci Please test Linux

[ahoppen]

@swift-ci Please test

[ahoppen]

@swift-ci Please test