Improve MWCC class unique ID normalization; normalize on the demangled name - #411

Fixes three issues with MWCC class unique ID normalization for symbol pairing:

• Previously only anonymous classes were supported, now named classes are also supported
• Multiple unique IDs within a single symbol are now all normalized (e.g. for function parameters)
• The length of the unique ID being different on the left vs right is now ignored (previously, $132966 and $55068 would not have been recognized as the same)

The third issue is difficult to fix on the mangled name, as the length of the name is included within the symbol, so I switched it to normalize and compare on the demangled version of the name, which does not include these lengths. This is inconsistent with how the mangled name is usually used for comparisons in all other cases in objdiff. However, it's a simpler solution than trying to fix the mangled name lengths, as that requires duplicating a large chunk of demangler logic within objdiff (#410).

Only downside I'm aware of here is that if the demangler ever failed to work on one of these symbols for some reason, the normalization would also not happen either then.