SLIMMED 0.9 — same corrected engine, redundant archives removed.
Removed from the original 0.9 upload (were ~8MB of baggage, not engine code):
  - 11 nested *.zip snapshots (v0.8.zip, v0_8_fixed (1..6).zip, v0.9/*.zip ...)
  - extracted/ and extracted6/ analysis dirs (unpacked duplicates of the above,
    each with their own tb_gguf.c + compiled binaries)
  - __pycache__, obj/, *.o, and the prebuilt Windows tb_infer.exe
Kept: full engine source (layer0..5, src, include), corrected layer4/tb_gguf.c
(+ .orig), build_cpu_verified.sh, and the rebuilt Linux bin/tb_infer_cpu_verified.
Verified: builds clean from this tree; runtime self-test = TRAILBLAZE ... PASS;
Q4_K byte-exact; 0 buggy mappings.
Note: this trims 0.9's own archived history. The corrected extracted*/ copies
still live in the full Trailblaze-0.9-q4k-corrected.zip if you need them.
