ARM64: Fix fdlibm compilation flags.

Currently fdlibm is compiled using the default C standard (gnu99).
This specifies a relaxed floating point model that enables fused
operations such as fused multiply-add (-ffp-contract=fast).

Since fdlibm is used to compute StrictMath operations this is too
permissive on targets that have support for fused operations such as
ARM64.

bug: 17663117
Signed-off-by: Serban Constantinescu <serban.constantinescu@arm.com>

(cherry picked from commit 0eb59337)

Change-Id: I713acb6b273787ae70ce1d664bea012b0a347ccb