Allow HSW stages to detect FMAs opportunistically.

We used to build the .S files with -ffp-contract=fast for this, and
adding it here to the HSW opts will have the equivalent effect.  None of
the other x86 targets support FMAs.

In many key places we do call _mm256_fmadd_ps explicitly, but in several
other stages we have been relying on Clang to sniff out opportunities,
especially for fused-multiply-subtracts, which we're just not very good
at seeing in our heads.

Bug: 821012
Change-Id: I2767cbdb7151d60ea7553e83f51b28ff9340eb94
Reviewed-on: https://chromium-review.googlesource.com/963141Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Mike Klein <mtklein@chromium.org>
Cr-Commit-Position: refs/heads/master@{#543262}

skia/BUILD.gn

@@ -648,6 +648,7 @@ if (current_cpu == "x86" || current_cpu == "x64") {
     }
     if (!is_win) {
       cflags = [
+        "-ffp-contract=fast",
         "-mavx2",
         "-mbmi",
         "-mbmi2",