Add support for inlinee lines

[mitsuhiko]

Refactors symbol parsing, introduces support for inlinees and parsing of their binary annotations into line records.

Breaking Changes:

• Symbol::name has been removed, since not all symbols have a name. Instead, every symbol with a name now exposes a name field. To provide uniform access to the name, SymbolData::name now exposes an Option<RawString> after parsing. (NB: This felt more robust than maintaining a manual list of offsets)
• SymbolData::Public and SymbolData::Data have been stripped of their *Symbol prefix, as this is redundant and inconsistent.
• Many symbol kinds, most notablySymbolData::CompileFlags, have been fixed with stronger typing.
• TypeIndex is now a newtype around u32, and so is IdIndex. This is in preparation of supporting the IPI stream. I'm yet undecided whether to expose the IPI in a separate type or via directly TypeFinder.

Public Changes:

• Symbol::starts_scope and Symbol::ends_scope now indicate which symbols can contain sub-symbols. This helps to deal with a hierarchy of symbols, such as building a tree.
• Every Symbol now exposes an index. Also, some symbols may refer to others via that index (e.g. inline sites point to their parent and their end). SymbolTable, ModuleInfo and SymbolIter now allow jumping to symbols by their index.
• Inline Sites are now supported via SymbolData::InlineSite(InlineSiteSymbol). It exposes an iterator over its binary annotations, which can be used to evaluate the line program of the inline site.
• LineProgram::inlinees allows iteration over inlinees in a module. Each inlinee has an inlinee_lines function that returns an iterator over line records evaluated from the binary annotations in the inline site symbol.

Internal Changes:

• Pread is now implemented for more types, including: Variant, CPUType, SourceLanguage.
• ParseBuffer can now parse any type that implements TryFromCtx, not just the ones that use scroll::Endian as context. There's also a new method parse_with to pass a custom parse context.
• SectionOffset types now implement AddAssign which has the same semantics as Add.

Fixes #14

[jan-auer]

This is not yet complete, but will contain a full refactor of Symbols to fix #14. Additionally, this helps to expose binary annotations more neatly.

[jan-auer]

Edit: Updated PR description.

[jan-auer]

@willglynn This is ready for review now. See #47 (comment) for a summary of all changes.

[jan-auer]

When working on the next feature (Cross Module Imports/Exports), I realized that inlinees are probably better placed on the module info directly than the line program.

[willglynn]

LGTM. Thanks! 🙌