SIMDTypes : Cleaner bit casting implentation and more Vc:: isolation

A few small clean ups

• A better float -> int type conversion (for bit casting in FastMaths) that avoids the header having to explicitly know all possible types.
• Define some types for memory alignment that avoids downstream code having to explicitly refer to Vc:: types. Currently implementation is still based on Vc::, but this allows that to change, for instance when using VE by default.
• Small improvement to Truncate.h to handle case when precision is higher than possible (fallbacks to no truncation).
• minor tweaks to test applications.