Fix bracket parsing logic for objdump labels

[connorsimms]

Resolves compiler-explorer/compiler-explorer#8809

This PR fixes an issue where the asm parser fails to correctly highlight/filter functions containing nested or imbalanced brackets such as
call 410145 <void a<int, int>()>
and
https://godbolt.org/z/3367rMndr.

The original logic used a forward-pass that overwrote the label start position at every < and terminated at the first > or +. In certain cases with nested/imbalanced brackets (e.g. operator<<), the complete label is not captured.

Instead of tracking label-state character-by-character, this version waits until the end of the instruction, finds the rightmost > character, and then scans backwards for a < preceded by a space " <" (unless the < is the first character).

This is safe because it is consistent with objdump formatting logic,
GNU Binutils objdump.c:

objdump_print_addr_with_sym (// ...
{
  if (!no_addresses)
    {
      objdump_print_value (vma, inf, skip_zeroes);
      (*inf->fprintf_styled_func) (inf->stream, dis_style_text, " ");
    }
// ...
(*inf->fprintf_styled_func) (inf->stream, dis_style_text,"<");

Note on tests:
I added ce-bug-8809.approved.txt. I also updated ce-bug-3963.approved.txt. The original parser was prematurely capturing char in <std::ctype<char>::_M_widen_init() const@plt> and returning char as the label. The new logic captures the entire name and allows the @plt filter to work properly.

[connorsimms]

Update: I pushed an additional commit to harden the backwards-scan condition.

I realized that scanning for " <" is a necessary condition, but it is not sufficient. For example, a templated operator< looks like this in the objdump output:
call 401128 <bool operator< <Dummy>(Dummy const&, Dummy const&)>. If we only look for a space then it would truncate the label at <Dummy>. I updated the logic to check that < is also preceded by a hex digit.

[connorsimms]

Actually, since objdump --no-addresses gets rid of the hex values altogether, it is better to just find the left most " <" instead of breaking early. I removed the hex check and the break statement. This should handle the edge cases mentioned above.