Fix: main cannot build the ROM -- accept a vptr store already in the ROM's vtable convention

[andrewboudreau]

main cannot build the ROM

python tools/rombuild.py -j 16 on baa3d97ee stops at step 3:

[3/6] mwccarm: 11077 enrolled source file(s), -j16
!! mwccarm failed (exit 1)
src/actors/daObjPathLift_c.cpp: isolate: compiler-only deadstrip refused:
  _ZTV10dBgActor_c: unexpected reloc type=2 addend=0

Reproduced on a clean detached checkout of origin/main. #1987 and #1979 are each
fine on their own; the break is in the pair. #1987 added a shape survey for the RTTI
relocations a promoted TU's surviving code makes, and #1979's promoted
ov100/daObjPathLift_c is the first source on main to exercise it.

The cause

There are two legitimate spellings of a vptr store, and they differ by exactly one
vtable preamble. The survey accepted only one of them.

mwcc's _ZTV<C> addresses the vtable object, so a store against a definition in
the same object carries the 8-byte preamble skip, and has to lose it once that
definition is externalised — that is the case #1987 was written for, and it is right.

But a source may equally declare the ROM symbol itself. daObjPathLift_c.cpp:79:

extern int _ZTV10dBgActor_c[];
...
*(void***)p = (void**)_ZTV10dBgActor_c;

That is symbols.txt's convention, where the symbol is the slot array. The store
relocates with addend 0, there is nothing to correct, and _apply already leaves it
alone — its rewrite is keyed on addend >= VTABLE_PREAMBLE. The survey refused it
anyway, so the first promoted TU to inherit a base's vtable rather than own one took
the whole build down.

This is not specific to daObjPathLift_c. Every promoted TU whose class inherits its
vptr — the common case — spells it this way.

The fix

Accept both spellings. Anything strictly between them is still refused rather than
guessed at, which is the point of having a survey at all.

shape = ((addend == 0 or addend >= VTABLE_PREAMBLE)
         if target.name.startswith("_ZTV") else addend == 0)

Two tests: one builds a class whose destructor stores an extern _ZTVBase, asserts the
reference relocates at 0 and comes back out untouched; the other retargets a known-good
addend to 4 and asserts the refusal still fires.

Verification

• python tools/rombuild.py -j 16 — 106/106 exact, 100.000000% of compared bytes,
  PASS; 446 verified / 239 partial / 15 differ from 7,798 object records.
• pytest tools/test_objisolate.py tools/test_rombuild.py tools/test_tubuild.py —
  99 passed.

🤖 Generated with Claude Code

[tangos-validator]

✅ PR validation — Passed

noverify: no source/build-data changes in this PR

_{Each changed src/*.c|*.cpp is compiled and its relocated bytes compared to the binary data on a private build box. Passing requires every changed file to reproduce the ROM byte-for-byte with correct relocation targets — this catches WRONG-DEST relocations and non-reproducing near-misses that ledger-scoped linkcheck skips.}