lib: align source before using optimized implementation

If the source is at the boundary of the VMA, loading one word at a
time can cause an alignment fault when the adjacent VMA is IO mapped.
Do byte-by-byte copy until source aligns to 8 bytes and then continue
with optimized version.

CRs-Fixed: 973724
Change-Id: I05e085597c58169fc6e275508a907029b9c7ec64
Signed-off-by: Sanrio Alvares <salvares@codeaurora.org>