qudaCloverInvert issues with MILC

[mathiaswagner]

Carleton reported issues with using the BiCGstab inverter from MILC using clover (qudaCloverInvert).
Points to look at include:

• Interface: hardcoded verbosity
• Interface: passing relative residual
• Interface: missing documentation of clover coefficient and input fields
• BiCG: bad convergence

I will post more details later. For now this is just a reminder to get the discussion from mails to github.

[mathiaswagner]

This discussion has been mainly done offline. It seems like the fixes done in the feature/milc_interface_cloverinvert and turning off GPU_COMMS at least make the runs agree with the results from the QOPQDP inverter.

To coordinate the work on that I created separate issues:
#231
#232

The hardcoded options in the milc_interface are tracked in #222 and the missing documentation for the interface in #233.

[mathiaswagner]

I hope all issues that showed up here are either solved or tracked in different issues now.