import h5py in the function call to filter numpy warnings (#37)

## Context
"special" binaries which flushes denormal results to zero triggers the
numpy warning for subnormal in every rank
after "import h5py". Neurodamus had a filter to ignore this type of
warning, but it is insufficient after the recent change of adding
"import h5py" at the top of replay.py. We have flooded warnings at the
beginning of the simulation before instantiating neuron. This PR changes
to import h5py in the function call to make the filter work again. It
also ensures that we are loading the right h5py version.
 
## Scope
In replay.py, import h5py in the function call instead of at the top.

## Testing
Existing tests

## Review
* [x] PR description is complete
* [x] Coding style (imports, function length, New functions, classes or
files) are good
* [ ] Unit/Scientific test added
* [ ] Updated Readme, in-code, developer documentation

neurodamus/replay.py

@@ -3,7 +3,6 @@
 """
 from __future__ import absolute_import
 import os
-import h5py
 import logging
 import numpy
 from .utils.logging import log_verbose
@@ -61,6 +60,7 @@ def open_spike_file(self, filename, delay, population=None):
 
     @classmethod
     def _read_spikes_sonata(cls, filename, population):
+        import h5py
         spikes_file = h5py.File(filename, "r")
         # File should have been validated earlier
         spikes = spikes_file.get("spikes/" + population)