Don't overwrite default numpy bindings from liglow in record.io

[cdcapano]

Currently, if you import the record module in pycbc.io, it will cause numpy to start casting float and int to float32 and int32 respectively. The default in numpy is to cast float and int to 64 bit scalars. The cause of this is due to the numpy scTypeDict being updated with ligolw bindings in record.py. In ligolw, float is cast to float32 and int to int32. This patch fixes that by not including the float and int mappings from ligolw. This will cause single precision values to be cast to double when reading a ligolw xml file, but that should only result in some meaningless extra digits being added to values.

[cdcapano]

Changed to not include any bindings that numpy doesn't already define (so, adds double to the list of exclusions).

[spxiwh]

Looks good to me. Only stylistic comment is why to use variable names preceded by an underscore for the internal variables in the dict-comprehension?

[cdcapano]

@spxiwh I was thinking these variables might be left over in the io.record namespace, so preceded them with _ to make them hidden. But you're right, because they're in a dict comprehension they don't survive outside of it, so that wasn't necessary. It doesn't hurt though, and doesn't violate PEP8, so I'm just going to leave it as is.