gradflow topological charge measurement

[kostrzewa]

TODO

• Our OpenMP accumulators assume that only one quantity is accumulated, this needs to be generalised.
• It must be ensured that Q is correctly normalised.
• Output of Q must be enabled
• 1-to-1 cross-check of energy density measurement on realistic lattice to make sure that I didn't break anything

[kostrzewa]

@urbach: This version contains measurements of the communication overhead at g_debug_level >= 4, so we can check if it makes sense to implement two-row gauge compression for communication.

[kostrzewa]

Okay, it seems that just one more check is required and we'll have a working topo charge measurement.

[kostrzewa]

The code has now been validated on three gauge configurations against @uwenger's (thanks!) and only round-off differences were found, which likely arise due to our use of Kahan sums in the process and thread-local reductions.