refactor: remove H5 storage backend in favor of mmap bin

- Remove H5Store, H5Writer, save_h5/load_h5 and h5py dependency
- MmapStore (bin) is the sole pre-tokenized storage backend
- Move setup_logging after imports to fix E402 in __init__.py
- Clean up unused imports across test files
- Move inline test imports to file top

astrai/__init__.py

@@ -4,30 +4,6 @@ __author__ = "ViperEkura"
 import logging
 import os
 
-
-def setup_logging(level: str = "INFO"):
-    """Attach a handler to the ``astrai`` logger (only, not root).
-
-    Call once per process, e.g. at the top of CLI scripts.
-    Set ``ASTR_LOG_LEVEL`` to override the default ``INFO``.
-    """
-    _logger = logging.getLogger("astrai")
-    if _logger.handlers:
-        return
-    _level = getattr(
-        logging, os.environ.get("ASTR_LOG_LEVEL", level).upper(), logging.INFO
-    )
-    _logger.setLevel(_level)
-    _handler = logging.StreamHandler()
-    _handler.setFormatter(
-        logging.Formatter(
-            "%(asctime)s | %(levelname)-7s | %(name)s | %(message)s",
-            datefmt="%Y-%m-%d %H:%M:%S",
-        )
-    )
-    _logger.addHandler(_handler)
-
-
 from astrai.config import (
     AutoRegressiveLMConfig,
     BaseModelConfig,
@@ -80,6 +56,30 @@ from astrai.trainer import (
     Trainer,
 )
 
+
+def setup_logging(level: str = "INFO"):
+    """Attach a handler to the ``astrai`` logger (only, not root).
+
+    Call once per process, e.g. at the top of CLI scripts.
+    Set ``ASTR_LOG_LEVEL`` to override the default ``INFO``.
+    """
+    _logger = logging.getLogger("astrai")
+    if _logger.handlers:
+        return
+    _level = getattr(
+        logging, os.environ.get("ASTR_LOG_LEVEL", level).upper(), logging.INFO
+    )
+    _logger.setLevel(_level)
+    _handler = logging.StreamHandler()
+    _handler.setFormatter(
+        logging.Formatter(
+            "%(asctime)s | %(levelname)-7s | %(name)s | %(message)s",
+            datefmt="%Y-%m-%d %H:%M:%S",
+        )
+    )
+    _logger.addHandler(_handler)
+
+
 __all__ = [
     "AutoRegressiveLM",
     "AutoRegressiveLMConfig",