shader_recompiler: Add subpixel offset for correct rounding at ImageGather

On AMD a subpixel offset of 1/512 of the texel size is applied to the texture coordinates at a ImageGather call to ensure the rounding at the texel centers is done the same way as in Maxwell or other Nvidia architectures.
See https://www.reedbeta.com/blog/texture-gathers-and-coordinate-precision/ for more details why this might be necessary.

This should fix shadow artifacts at object edges in Zelda: Breath of the Wild (#9957, #6956).

src/video_core/renderer_opengl/gl_device.cpp

@@ -169,6 +169,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
     has_draw_texture = GLAD_GL_NV_draw_texture;
     warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel;
     need_fastmath_off = is_nvidia;
+    need_gather_subpixel_offset = is_amd;
     can_report_memory = GLAD_GL_NVX_gpu_memory_info;
 
     // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive

src/video_core/renderer_opengl/gl_device.h

@@ -160,6 +160,10 @@ public:
         return need_fastmath_off;
     }
 
+    bool NeedsGatherSubpixelOffset() const {
+        return need_gather_subpixel_offset;
+    }
+
     bool HasCbufFtouBug() const {
         return has_cbuf_ftou_bug;
     }
@@ -225,6 +229,7 @@ private:
     bool has_draw_texture{};
     bool warp_size_potentially_larger_than_guest{};
     bool need_fastmath_off{};
+    bool need_gather_subpixel_offset{};
     bool has_cbuf_ftou_bug{};
     bool has_bool_ref_bug{};
     bool can_report_memory{};

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -218,6 +218,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
           .lower_left_origin_mode = true,
           .need_declared_frag_colors = true,
           .need_fastmath_off = device.NeedsFastmathOff(),
+          .need_gather_subpixel_offset = device.NeedsGatherSubpixelOffset(),
 
           .has_broken_spirv_clamp = true,
           .has_broken_unsigned_image_offsets = true,