Debug version now builds OK and all_source compiles

[LagoLunatic]

All ShieldD errors fixed:

$ ninja
[1/1] PROGRESS
Progress:
  All: 9.65% matched, 0.00% linked (1 / 3201 files)
    Code: 1422620 / 14741676 bytes (20460 / 64457 functions)
    Data: 162968 / 4133020 bytes (3.94%)
  DOL: 13.38% matched, 0.00% linked (0 / 875 files)
    Code: 840048 / 6279040 bytes (11196 / 31033 functions)
    Data: 150432 / 1971544 bytes (7.63%)
  Modules: 6.88% matched, 0.00% linked (1 / 2326 files)
    Code: 582572 / 8462636 bytes (9264 / 33424 functions)
    Data: 12536 / 2161476 bytes (0.58%)
  TP Game Code: 7.50% matched, 0.00% linked (1 / 923 files)
    Code: 903952 / 12052876 bytes (12698 / 45761 functions)
    Data: 34248 / 3081100 bytes (1.11%)
  Core Engine: 24.43% matched, 0.00% linked (0 / 92 files)
    Code: 110668 / 453028 bytes (1332 / 2940 functions)
    Data: 25372 / 271100 bytes (9.36%)
  SDK: 55.56% matched, 0.00% linked (0 / 1164 files)
    Code: 103324 / 185972 bytes (2679 / 3168 functions)
    Data: 4712 / 11112 bytes (42.40%)
  Third Party: 29.78% matched, 0.00% linked (0 / 258 files)
    Code: 304676 / 1023000 bytes (3751 / 6955 functions)
    Data: 98636 / 251456 bytes (39.23%)

You could add ShieldD to github CI now if you want to track progress and catch future build errors, note that if you do the .dol needed in the build repo must be converted via this dtk build: https://github.com/encounter/decomp-toolkit/actions/runs/14001786651
By using the dtk elf2dol command on the debug .alf.

No match% regressions for retail (only difference is two funcs in vrbox2 increased in % after I messed with them in debug).

I didn't link anything besides f_pc_profile_lst because there are some weird issues that aren't resolved yet.

For example, d_a_andsw matches 100% now, but the linker isn't stripping out weak functions that are used in the TU, only ones that are unused. I tried various linker flags (e.g. code merging) but nothing seemed to help.

Also name mangling issues for weak functions in m_Do_mtx:
void* JGeometry::TVec3<float>::TVec3<float>(float, float, float)
vs
JGeometry::TVec3<float>::TVec3(float, float, float)