Omit all optional SkOpts code when optimizing for size

This is a blunt way to reduce code-size, but the actual impact (and the
traded-off amount of performance regression) are hard to quantify:

You will now get one copy of all the SkOpts code, compiled using
whatever architecture you're targeting. If you set the minimum
architecture high enough, then the performance loss is minimal.
If not, you could be running fairly slow scalar code in many places.

Similarly, ARM builds never had runtime selected NEON code, so you only
really save on the optional CRC32 optimizations. X64 builds potentially
cut out many more permutations of code.

Bug: skia:13657
Change-Id: I439d1f50bc0ab53e0a83c689fe026ff862292241
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/580962
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
7 files changed