sanitycheck: parallelize binary size calculations

Instead of running nm/objdump and parsing the results one testcase
at a time, do these in parallel to speed up sanity check a bit.

The sanitycheck now requires at least Python 3.2 to run because of
the usage of concurrent.futures.

This reduces build time (--all -b) from 4:30 to 3:08 on 24C/48T
Xeon-E5.

Change-Id: I8e7c1efb2f473c7f2b65658f8ed9a101ed091eea
Signed-off-by: Daniel Leung <daniel.leung@intel.com>