jemalloc: apply recommended optimizations

The Jemalloc project suggests using
-O3 and -funroll-loops for optimization.
This commit enables these flags by default
in order to improve the performance.

Test: Compiled arm, arm64, x86, and x86_64.
Test: Booted on arm and arm64.
Test: Passed full CTS suite run on arm64.

Change-Id: Ibb8305be459db2e670e335ab6eac333855654379
Signed-off-by: Alex Naidis <alex.naidis@linux.com>