DM: track a direct/indirect bit for each Sink too.

The decoding tests can now veto indirect sinks like pipe-8888.

This moves Sink type detection from automatic to explicit; I can't think of any
way to automatically differentiate pipe-8888 from 8888 based only on the
output.  (They should ideally be identical, after all.)

BUG=skia:4138

Review URL: https://codereview.chromium.org/1263113002
diff --git a/dm/DMSrcSink.cpp b/dm/DMSrcSink.cpp
index 0cd2ac9..d1d9be0 100644
--- a/dm/DMSrcSink.cpp
+++ b/dm/DMSrcSink.cpp
@@ -71,11 +71,12 @@
     , fScale(scale)
 {}
 
-bool CodecSrc::veto(SinkType type) const {
-    // No need to test decoding to non-raster backend.
+bool CodecSrc::veto(SinkFlags flags) const {
+    // No need to test decoding to non-raster or indirect backend.
     // TODO: Once we implement GPU paths (e.g. JPEG YUV), we should use a deferred decode to
     // let the GPU handle it.
-    return type != kRaster_SinkType;
+    return flags.type != SinkFlags::kRaster
+        || flags.approach != SinkFlags::kDirect;
 }
 
 Error CodecSrc::draw(SkCanvas* canvas) const {
@@ -460,10 +461,11 @@
 
 ImageSrc::ImageSrc(Path path, int divisor) : fPath(path), fDivisor(divisor) {}
 
-bool ImageSrc::veto(SinkType type) const {
-    // No need to test decoding to non-raster backend.
+bool ImageSrc::veto(SinkFlags flags) const {
+    // No need to test decoding to non-raster or indirect backend.
     // TODO: Instead, use lazy decoding to allow the GPU to handle cases like YUV.
-    return type != kRaster_SinkType;
+    return flags.type != SinkFlags::kRaster
+        || flags.approach != SinkFlags::kDirect;
 }
 
 Error ImageSrc::draw(SkCanvas* canvas) const {