Optimizes Blosc2 MLArray data loading

Refactors the in-memory data loading process for Blosc2 MLArrays from disk. Previously, data was loaded via an intermediate cframe conversion. Now, an empty Blosc2 array is created in memory, and data chunks along with metadata are explicitly copied.

This ensures a more direct and controlled preparation of the array in memory, better supporting optimizations for subsequent operations.

Author: Karol-G

mlarray/mlarray.py

@@ -1540,8 +1540,19 @@ def _load(
         self.support_metadata = True
         dparams = MLArray._resolve_dparams(dparams)
         ondisk = blosc2.open(str(filepath), dparams=dparams, mode="r")
-        cframe = ondisk.to_cframe()
-        self._store = blosc2.ndarray_from_cframe(cframe, copy=True)
+        in_memory = blosc2.empty(
+            ondisk.shape,
+            dtype=ondisk.dtype,
+            chunks=ondisk.chunks,
+            blocks=ondisk.blocks,
+            cparams=ondisk.schunk.cparams,
+            dparams=ondisk.schunk.dparams,
+        )
+        for i in range(ondisk.schunk.nchunks):
+            in_memory.schunk.update_chunk(i, ondisk.schunk.get_chunk(i))
+        for key, value in ondisk.schunk.vlmeta.getall().items():
+            in_memory.schunk.vlmeta[key] = value
+        self._store = in_memory
         self._backend = "blosc2"
         self.mode = None
         self.mmap_mode = None