RF: use new cached append for multi-seq read

Use the new cached append to replace the coroutines for the read across
multiple array sequences in ``create_arraysequences_from_generator``.

Before this change (using coroutines):

```
Old: Loaded 5,000 streamlines in   2.85

New: Loaded 5,000 streamlines in    3.6
Speedup of 0.79
Old: Loaded 5,000 streamlines with scalars in   5.16
New: Loaded 5,000 streamlines with scalars in   7.13
Speedup of 0.723703
```

After this change (using cached append):

```
Old: Loaded 5,000 streamlines in   3.21

New: Loaded 5,000 streamlines in    3.9
Speedup of 0.82
Old: Loaded 5,000 streamlines with scalars in   5.21
New: Loaded 5,000 streamlines with scalars in   7.16
Speedup of 0.727654
```

This seems to be well within run-to-run measurement error.

Author: matthew-brett

nibabel/streamlines/array_sequence.py

@@ -444,17 +444,11 @@ def create_arraysequences_from_generator(gen, n):
         Number of :class:`ArraySequences` object to create.
     """
     seqs = [ArraySequence() for _ in range(n)]
-    coroutines = [seq._extend_using_coroutine() for seq in seqs]
-
-    for coroutine in coroutines:
-        coroutine.send(None)
-
     for data in gen:
-        for i, coroutine in enumerate(coroutines):
+        for i, seq in enumerate(seqs):
             if data[i].nbytes > 0:
-                coroutine.send(data[i])
-
-    for coroutine in coroutines:
-        coroutine.close()
+                seq.append(data[i], cache_build=True)
 
+    for seq in seqs:
+        seq.finalize_append()
     return seqs