Fix handling of numexpr.cpu in print_versions()

[pinotree]

There are a couple of issues in the way numexpr.cpu.info is used in print_versions():

1. all the subclasses of CPUInfoBase have a class attribute info, whereas CPUInfoBase does not; this means that, in case there is no CPU implementation for an OS, self.info returns a function (because of CPUInfoBase.__getattr__()) which is not subscriptable
2. only few subclasses of CPUInfoBase (e.g. LinuxCPUInfo, Win32CPUInfo) implement info as list of dicts for each CPU, whereas the rest of the implementations use a dict with attributes

Since info seems to be mostly an internal detail, then adapt print_versions() to be a little more flexible:

• do not try to query attributes in case the actual CPU implementation is CPUInfoBase: it is not implemented for the current OS, so there is nothing to query anyway (fixes (1))
• get the first element of info only in case it is a list (fixes (2))
• drop the catch of KeyError, which should not happen anymore now