Refactor compiler diagnostic data structures

[bamarsha]

• Diagnostics need additional information to display a diagnostic message, but each diagnostic is an enum item, which doesn't allow additional data. The information is stored as a string sequence which is error-prone and causes confusion since each string is referred to by its index instead of a name.
• Diagnostic enums are grouped by severity, but severity should be a separate configurable property of each diagnostic, not hard-coded into its name. This also means that diagnostic codes are duplicated across severities, which is confusing.
• Diagnostic items require a range, but not all diagnostics have a range, which causes a default range to be used as a fallback throughout the compiler whenever a diagnostic is created. Diagnostics should make the range optional and use a standard default value to reduce boilerplate/code duplication.

[bamarsha]

An example of where the code for diagnostic messages starts to get really messy and hard to maintain is when we want to make the diagnostic message a bit smarter/adaptable to the context:

https://github.com/microsoft/qsharp-compiler/blob/29d10588c2b85a5ffe8e4243eb94cd96f1a3adbc/src/QsCompiler/DataStructures/Diagnostics.fs#L474-L481

(I don't think it's right to create separate diagnostics for the two messages, because they really are the same diagnostic, just displayed differently depending on context.)