GameTank (Clyde Shaffer's open W65C02S console) was claimed as a
rebuildable-project platform in the docs but disasm({target:'project'})
threw 'no regions planned for gametank' — the claim was false. Rather
than downgrade the count, wire it in so the docs become true: it's a
flat 32KB EEPROM cart at $8000 (vectors at $FFFA) on the 6502-family
da65/ca65 path, so it drops straight into planRegions + CPU_FAMILY.

- planRegions: gametank branch (one flat region, org = 0x10000 - size).
- CPU_FAMILY: gametank -> 6502 (W65C02 is a 65C02 superset; 65C02-only
  opcodes floor to .byte, still byte-exact).
- sniffPlatformFromPath: .gtr -> gametank. PLATFORM_VIRTUAL_EXT: .gtr.
- Verified byte-identical round-trip; test/reassemble-roundtrip.test.js
  gains a GameTank case (9/9). All 15 classic platforms now confirmed
  byte-exact via build({output:'reassemble'}).

Docs sweep for the reassemble feature (removes now-stale 'no one-call
rebuild / run native as/ld/objcopy yourself' claims, corrects 14->15):
- README.md, romdevtools README/AGENTS.md, CHANGELOG.md.
- disasm tool description (the agent-facing one) now leads with the
  one-call reassemble rebuild.
- ROMHACKING_PLAYBOOK.md §7b rewritten: the 'two rebuild tiers' split is
  replaced by 'one call, every platform'; rebuild.json demoted to the
  cc65-native alternate.
- Per-platform MENTAL_MODELs: snes, gba, sms, msx, atari2600 (the ones
  that described rebuild as native-recipe-only).