add Q14x2 type to SkVM

This is a 32-bit pair of Q14 (signed 1.14 fixed point) values.

Q14 * Q14 is (x*y + 0x2000)>>14 but we'll do it as
((x*y + 0x4000)>>15)<<1 to allow use of vpmulhrsw.

Somewhat awkwardly this approximate math does not preserve x*1.0 == x,
e.g.  0x0001 * 0x4000 = 0x0002, not 0x0001.  I'm unsure if we'll care.
x*0.0 == 0.0 does always hold, and 1.0*1.0 == 1.0.

Most other operations are normal 16-bit operations.

TODO:
    - implmement in interpreter
    - unit test
    - constant propgation and strength reduction
    - implement in JIT
    - effect/blitter hooks enough to demo
    - thorough effect/blitter support

Change-Id: Ic714fc16b6530259b36300bd042bbfd5a57a663b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/317149
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
3 files changed