E.g. in gpu_surface_gl.cc, we should be setting the byte size resource limit based on physically available memory, perhaps something like:

  context_->setResourceCacheLimits(kGrCacheMaxCount, std::min(kGrCacheMaxByteSize, fml::GetAvailablePhysicalMemory() / 3);

The current value is too high for low end devices with 1gb of ram or less. This will also requiring some kind of platform independent way of determining avialble physical memory (or something meaningful for a given platform anyway).