Make CPU coverage bot a bit faster

Adding -O1 and reducing the shards to cut down on overhead
in post-production.  This seems to save us a few minutes
(e.g. ~26 -> ~23 minutes).

The optimized code has about 600 fewer reachable lines of code
(according to the Coverage measurement), which is acceptable
given our tests run on -O1 or better.

Experiments (in this review) show that O3 is faster than O1, 
but not significantly enough where we want to deviate from 
our normal Debug bots:
Control: 73 minutes
-O1: 59 minutes
-O2: 60 minutes
-O3: 50 minutes

Bug: skia:
NOTRY=true
Change-Id: I33344c1cd2408373004d010e36ce27d6aa03deb2
Reviewed-on: https://skia-review.googlesource.com/65503
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Kevin Lubick <kjlubick@google.com>
6 files changed