quantcpp 0.9.0: KV compression ON by default in Python bindings

BREAKTHROUGH: kv_compress=1 was never broken in quant.h — the v0.8.1
abort was caused by the libc.free() cross-heap bug (fixed in v0.8.2
via quant_free_string), not by the UNIFORM_4B KV path. We isolated
the wrong variable because kv_compress=0 AND skip-free were changed
simultaneously in the v0.8.1 hotfix.

Verified in standalone C AND Python ctypes: kv_compress=1 (UNIFORM_4B)
works cleanly on SmolLM2-135M with quant_free_string. This is honest
correction #8: "we disabled a working feature because of incorrect
root cause analysis."

Changes:
  - kv_compress default restored to 1 (was 0 since v0.8.1)
  - kv_compress warning/fallback guard removed
  - Version bumped to 0.9.0 (major: KV compression is now the default
    experience for all pip users)

The headline value proposition now flows through both distribution
channels identically:
  CLI:    quant model.gguf -k turbo_kv_4b     → 7x KV compression
  Python: Model("model.gguf")                  → 4-bit KV compression

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: unamedkr

bindings/python/pyproject.toml

@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "quantcpp"
-version = "0.8.3"
+version = "0.9.0"
 description = "Single-header LLM inference engine with KV cache compression (7× compression at fp32 parity)"
 readme = "README.md"
 license = { text = "Apache-2.0" }

bindings/python/quantcpp/__init__.py

@@ -19,7 +19,7 @@
     from importlib.metadata import version as _pkg_version
     __version__ = _pkg_version("quantcpp")
 except Exception:
-    __version__ = "0.8.3"  # fallback for editable / source-tree imports
+    __version__ = "0.9.0"  # fallback for editable / source-tree imports
 
 import os
 import sys
@@ -181,28 +181,8 @@ def __init__(
         top_p: float = 0.9,
         max_tokens: int = 256,
         n_threads: int = 4,
-        kv_compress: int = 0,
+        kv_compress: int = 1,
     ):
-        """
-        .. note::
-           ``kv_compress=1`` and ``kv_compress=2`` are temporarily disabled in
-           the Python bindings (v0.8.x) — the bundled ``quant.h`` single
-           header carries an older KV compression path that aborts on Llama
-           architectures. The CLI ``quant`` binary uses the multi-file engine
-           and works with all KV types. KV compression will be re-enabled in
-           the bindings once ``quant.h`` is re-generated against the v0.8.0+
-           tree (tracked as v0.8.1: WASM SIMD / un-stub turbo_kv).
-        """
-        if kv_compress not in (0,):
-            import warnings
-            warnings.warn(
-                "kv_compress != 0 is not supported in the Python bindings of "
-                "quantcpp 0.8.x — falling back to kv_compress=0. Use the CLI "
-                "binary for KV compression until v0.8.2.",
-                RuntimeWarning,
-                stacklevel=2,
-            )
-            kv_compress = 0
         if not os.path.isfile(path):
             raise FileNotFoundError(f"Model file not found: {path}")