DLOG instead of DCHECK destruction during write access

This is motivated by our fuzzing of RasterDecoder. The fuzzer
can issue a BeginRasterCHROMIUM without a matching EndRasterCHROMIUM.
The DCHECK was causing clusterfuzz to think it found an legit error.

It is possible for a renderer to crash between BeginRaster and EndRaster
as well.

Bug: 940190
Change-Id: Ia260b626ce516a700fc1159e4be33c7adb02740c
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1518144
Commit-Queue: Jonathan Backer <backer@chromium.org>
Reviewed-by: Eric Karl <ericrk@chromium.org>
Reviewed-by: enne <enne@chromium.org>
Auto-Submit: Jonathan Backer <backer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#640365}

gpu/command_buffer/service/shared_image_backing_factory_gl_texture.cc

@@ -273,7 +273,12 @@ class SharedImageRepresentationSkiaImpl : public SharedImageRepresentationSkia {
 #endif
   }
 
-  ~SharedImageRepresentationSkiaImpl() override { DCHECK(!write_surface_); }
+  ~SharedImageRepresentationSkiaImpl() override {
+    if (write_surface_) {
+      DLOG(ERROR) << "SharedImageRepresentationSkia was destroyed while still "
+                  << "open for write access.";
+    }
+  }
 
   sk_sp<SkSurface> BeginWriteAccess(
       GrContext* gr_context,