[io] write_spec_to_h5 can't handle empty scans about silx

[PiRK]

I found several problems in the SpecFile C library:

• SfAllLabels returns a non zero error status when parsing a #L line in an empty scan. My workaround is to catch the generated IndexError and manually parsing the #L line (splitting on double-spaces)
• SfData causes a Segmentation Fault when trying to retrieve data from an empty scan. My workaround is to look for a #C line containing the string aborted after 0 points in the scan header. I'm not sure if this is a reliable safeguard, maybe it depends on the SPEC version used to write the file. An alternative method is to try to read a single data column with SfDataColumnByName (which doesn't cause a segfault) before calling SfData, but I suspect this would affect performance more than testing the scan header for a #C line.

[PiRK]

SfData doesn't directly cause the segmentation fault, but it doesn't seem to allocate memory for data_info in case there are no data points. Accessing data_info[0] caused the segfault. I fixed the bug in a cleaner way by checking <long>data_info != 0 before accessing the array.