feat: core profiles, data_dirs buildbot, cross_ref fix

profiles: amiberry (new), amiarcadia, atari800, azahar, b2,
bk, blastem, bluemsx, freeintv updated with source refs,
upstream field, mode field, data_directories.

_data_dirs.yml: buildbot source for retroarch platforms,
strip_components for nested ZIPs, freeintv-overlays fixed.

cross_reference.py: data_directories-aware gap analysis,
suppresses false gaps when emulator+platform share refs.

refresh_data_dirs.py: ZIP strip_components support,
for_platforms filter, ETag freshness for buildbot.

scraper: bluemsx single ref, freeintv overlays injection.
generate_pack.py: warning on missing data directory cache.

emulators/freeintv_ts_overlay.yml

@@ -26,7 +26,10 @@ notes: |
 
   ROM-specific overlays are optional 370x600 PNG files placed by the user in
   system/freeintv_overlays/<rom_name>.png. These are cosmetic per-game keypad
-  images and are not BIOS files.
+  images loaded via stbi_load() (src/libretro.c:273). Core falls back to
+  embedded default_keypad_image.h if no per-game overlay found.
+  90 overlays available in Assets/Overlays.zip in the core repo.
+  Not on RetroArch buildbot — sourced from core repo via _data_dirs.yml.
 
   BIOS loading is unchanged from the base FreeIntv core. retro_load_game()
   (src/libretro.c:1169-1174) loads exec.bin and grom.bin from the system
@@ -34,6 +37,10 @@ notes: |
 
 bios_identical_to: freeintv
 
+data_directories:
+  - ref: freeintv-overlays
+    destination: freeintv_overlays
+
 files:
   - name: "exec.bin"
     system: intellivision