* fix develop build issue

* fix google style

* cpplint check only fluid

Use scalar implementation instead of neon implementation to avoid out of range memory access in the tail conv3x3.

"suggest braces around initialization of subobject [-Werror,-Wmissing-braces]"

Use template to deliver const argument instead, to remove the compiling error "argument to __builtin_neon_vgetq_lane_f32 must be a constant integer".