Optimize decompress_offset_alignr_cycle for Offset 25 (Shift 7) (#381)

Optimized the decompression hot path for Offset 25 by specializing `decompress_offset_alignr_cycle` for `SHIFT=7`.
The loop was unrolled to a stride of 96 bytes (6 vectors).
The serial dependency chain of `alignr` instructions was optimized by computing vectors `v_next2`, `v_next4`, and `v_next5` using accumulated shift constants (e.g., using shift 14 on `v_prev` and `v_align` directly instead of relying on `v_next1`). This reduces the dependency depth and increases instruction-level parallelism.

Performance Impact:
- Throughput for `Decompress offset25` improved by ~1.4% (from ~10.07 GiB/s to ~10.23 GiB/s).
- Verified correctness with `cargo test`.

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>

Author: 404Setup

src/decompress/x86.rs

@@ -442,6 +442,28 @@ unsafe fn decompress_offset_alignr_cycle<const SHIFT: i32>(
             _mm_storeu_si128(out_next.add(copied + 64) as *mut __m128i, v_next4);
             _mm_storeu_si128(out_next.add(copied + 80) as *mut __m128i, v_next5);
 
+            v_prev = v_next5;
+            v_align = v_next4;
+            copied += 96;
+        }
+    } else if SHIFT == 7 {
+        // Optimization for Offset 25 (Shift 7): Unroll loop to write 96 bytes (6 vectors) per iteration.
+        while copied + 96 <= length {
+            let v_next0 = _mm_alignr_epi8::<7>(v_prev, v_align);
+            let v_next1 = _mm_alignr_epi8::<7>(v_next0, v_prev);
+
+            let v_next2 = _mm_alignr_epi8::<14>(v_prev, v_align);
+            let v_next3 = _mm_alignr_epi8::<14>(v_next0, v_prev);
+            let v_next4 = _mm_alignr_epi8::<14>(v_next1, v_next0);
+            let v_next5 = _mm_alignr_epi8::<14>(v_next2, v_next1);
+
+            _mm_storeu_si128(out_next.add(copied) as *mut __m128i, v_next0);
+            _mm_storeu_si128(out_next.add(copied + 16) as *mut __m128i, v_next1);
+            _mm_storeu_si128(out_next.add(copied + 32) as *mut __m128i, v_next2);
+            _mm_storeu_si128(out_next.add(copied + 48) as *mut __m128i, v_next3);
+            _mm_storeu_si128(out_next.add(copied + 64) as *mut __m128i, v_next4);
+            _mm_storeu_si128(out_next.add(copied + 80) as *mut __m128i, v_next5);
+
             v_prev = v_next5;
             v_align = v_next4;
             copied += 96;