[tint][hlsl] Only polyfill trunc on FXC

The HLSL polyfill for trunc was added in:
https://dawn-review.googlesource.com/c/dawn/+/125420
to work around an FXC bug, but causes issues on newer Intel devices
using DXC, which have different flushing to zero behavior for subnormals
(see https://crbug.com/42251226), which then affects the polyfill of
remainder causing incorrect values for some f16 cases.

We can avoid this affecting remainder on Intel by limiting the polyfill
of trunc to FXC where it's actually needed.

As a follow up we should confirm whether there is a driver issue on
Intel that still needs to be covered by a new CTS test for use of
floor/ceil directly in WGSL.

Bug: 390458364
Change-Id: I79aec18393afa19f3c9047dd3ecaed0d6a6a6964
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/302655
Reviewed-by: dan sinclair <dsinclair@chromium.org>
Commit-Queue: Natalie Chouinard <chouinard@google.com>