Add openpyxl dependency

[dgdekoning]

When reading large .xlsx files, openpyxl is dramatically faster that xlrd because it does not have to read the entire file to determine the sheet names.

[coveralls]

Coverage increased (+0.02%) to 56.546% when pulling 9d33207 on openpyxl into cd34dd3 on master.

[dgdekoning]

Found out that 'on_demand' does nothing on xlsx files: https://stackoverflow.com/a/31187609.

Note that it is definitely possible to feed the openpyxl Workbook object into pandas.read_excel by using the engine='openpyxl' argument. There is a speed increase in doing so, but it is quite minor compared to the other two methods where openpyxl was used instead of xlrd.