Merge pull request #47 from AdaWorldAPI/claude/transcode-deepnsm-rust-oNa1Z

fix: f16 subnormal overflow + OpenChat 3.5 Q8_0 integration test

Fix signed arithmetic overflow in f16_to_f32 for subnormal exponents.
Add integration test that streams OpenChat 3.5 Q8_0 (7.7 GB) through
the bgz17 indexer → 42.6 MB output (679× overall compression).

Results: Attention 328×, FeedForward 920×, Embedding 3765×.
Peak RAM: 524 MB. Time: 185s. 226 tensors indexed, 65 skipped.

https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7

Author: AdaWorldAPI

src/hpc/gguf.rs

@@ -425,7 +425,8 @@ fn f16_to_f32(bits: u16) -> f32 {
             e -= 1;
         }
         m &= 0x3FF;
-        let f32_bits = (sign << 31) | (((127 - 15 + 1 + e as u32) & 0xFF) << 23) | (m << 13);
+        let f32_exp = (127i32 - 15 + 1 + e).max(0) as u32;
+        let f32_bits = (sign << 31) | ((f32_exp & 0xFF) << 23) | (m << 13);
         return f32::from_bits(f32_bits);
     }
     if exp == 31 {