Add actor profile reconstruction pilot

[andrewboudreau]

Summary

• add an evidence-only extractor and fixed 20-entry actor/process profile dataset
• export the complete registry as 391 logical IDs and 401 valid overlay-local candidate rows in TSV and JSON
• document class-initializer, profile-global, TU-filename, macro-pattern, field-layout, and lifecycle findings
• add dry-run rename maps with explicit collision and overlay-ambiguity refusals
• wire focused regression tests into the tooling test suite

No source, symbol map, delinks, or production profile rename is applied.

Main findings

• 14 sampled actor descriptors use the 0x1c SpawnInfo layout; 6 scene/process descriptors use a separate 0x08 layout
• all 391 registry entries have a valid descriptor interpretation whose +0x04 halfword equals the registry index; the SM64DS base constructor also uses that same halfword as behavior/execute order
• +0x06 supplies render/draw order; the actor extension directly supplies actor flags, clip-volume Y offset, clip radius, clip distance, and far distance
• a distinct group-flags descriptor field is not yet located and remains explicitly unresolved
• SM64DS exposes the same four do/pre/post lifecycle triplets as NSMBW: InitResources/create, CleanupResources/doDelete, Behavior/execute, and Render/draw are high-confidence role correspondences, not proven shared spellings
• 19/20 sampled factories have an ordinary construction-wrapper shape
• the complete candidate inventory has 42 colliding Class_classInit proposals among 314 distinct class-based spellings, so classInit is not a safe universal rename rule
• all 391 ROM profile IDs are unique, making g_profile_ROM_ID structurally consistent, subject to overlay-context checks
• TU stems are useful lineage candidates, but alphabetical ordering is not generally predictive and multi-stem TUs remain unresolved
• repeated layout and factory patterns support a macro-like architecture without recovering exact SM64DS macro syntax

Validation

• python tools/profile_reconstruction.py --check --no-refresh
• full explicit tooling suite: 535 tests passed, 3 skipped
• deterministic generated outputs
• documentation reference checks: no new dead or broken links
• port_refcheck: 405 references resolved
• pre-push: 97/97 translation units compiled

[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.}