Benchmark FPU disabled vs enabled

[cl91]

In the latest seL4 upstream lazy FPU restoration in context switches has been removed and threads always have FPU enabled. A thread can be marked as FPU disabled so context switching cost is considerably lower. We can compile the NT executive task and drivers (and base NT clients) without FPU and set FPU disabled for these tasks. We should benchmark the performance benefit/penalty of doing so.

[cl91]

There doesn't seem to be a ton of performance difference between FPU enabled vs FPU disabled (within 5% for disk IO). Still need more data to decide.