Add ET support for Lattice spin- and colour-traces

[RChrHill]

Currently, the functions TraceColour and TraceSpin, as well as the more generic TraceIndex ( Grid/lattice/Lattice_trace.h) are available for tracing over internal indices of lattice objects. However, these functions cannot take an expression template as an argument, requiring them to be called as e.g. TraceIndex<ColourIndex>(closure(a*b)) or for the expression to be assigned to a temporary first, forcing the full expression evaluation before the trace is performed.

This pull request adds expression-template-compatible traceColour and traceSpin overloads in the same spirit as trace, removing the need for a closure() and reducing the cost of a trace over individual indices. This is currently being used to obtain performance gains in a contraction-heavy in-development workflow.

This is a minimally-intrusive implementation that may not be ideal, in large part due to replicating the macros from Grid/lattice/Lattice_ET.h. To avoid this the macro #undefs at the bottom of Grid/lattice/Lattice_ET.h could be removed.
These operators can't live inside Grid/lattice/Lattice_ET.h because SpinIndex and ColourIndex are defined in Grid/qcd/QCD.h, and implementing a traceIndex op would require additional macro definitions to account for the Index template parameter -- which might be the better way to achieve this.

I will update the pull request with production GPU benchmarks in the next few days.