Expose Skia's Vulkan device-lost callback (#4601)

[ramezgerges]

Description

Exposes Skia's Vulkan device-lost callback (skgpu::VulkanBackendContext.fDeviceLostContext / fDeviceLostProc) on the C API so managed callers can react to VK_ERROR_DEVICE_LOST — the intended channel for detecting TDRs / driver resets.

Skia's own callback signature is (std::string description, std::vector<VkDeviceFaultAddressInfoEXT>, std::vector<VkDeviceFaultVendorInfoEXT>, std::vector<std::byte>). Not C-ABI-friendly. The shim adapts it into a gr_vk_device_lost_info_t POD struct that mirrors the full payload — description string + address-info array + vendor-info array + vendor binary blob. The Vk struct layouts are stable ABI, so the shim's gr_vk_device_fault_*_info_t mirror them field-for-field and reinterpret_cast the vector storage (no per-element copy). Static_assert guards the layout equivalence.

Same caller-owned bridge pattern as sk_graphite_shader_error_handler (_new returns an opaque handle; caller passes to backend context; caller deletes after Context destruction). No leaks.

SkiaSharp issue

Related to mono/SkiaSharp#4601

Required SkiaSharp PR

Requires mono/SkiaSharp#4655

Areas affected

• C API (include/c, src/c)
• Native dependency / DEPS
• Build (gn / build files)
• Upstream Skia merge or rebase
• Rendering output / behavior
• Other

Changes

C API

// New POD structs (mirror VkDeviceFault*InfoEXT layouts verbatim)
typedef struct { int32_t fAddressType; uint64_t fReportedAddress; uint64_t fAddressPrecision; } gr_vk_device_fault_address_info_t;
typedef struct { char fDescription[256]; uint64_t fVendorFaultCode; uint64_t fVendorFaultData; } gr_vk_device_fault_vendor_info_t;

typedef struct {
    const char* fDescription;
    const gr_vk_device_fault_address_info_t* fAddressInfos;  int32_t fAddressInfoCount;
    const gr_vk_device_fault_vendor_info_t*  fVendorInfos;   int32_t fVendorInfoCount;
    const void* fVendorBinaryData;                            size_t  fVendorBinaryDataSize;
} gr_vk_device_lost_info_t;

typedef void (*gr_vk_device_lost_proc)(void* userData, const gr_vk_device_lost_info_t* info);
typedef struct gr_vk_device_lost_handler_t gr_vk_device_lost_handler_t;

gr_vk_device_lost_handler_t* gr_vk_device_lost_handler_new(gr_vk_device_lost_proc proc, void* userData);
void                         gr_vk_device_lost_handler_delete(gr_vk_device_lost_handler_t* handler);

// gr_vk_backendcontext_t + sk_graphite_vk_backend_context_init_t both gain:
gr_vk_device_lost_handler_t* fDeviceLostHandler;   // nullable

Behavior

• Ganesh Vulkan + Graphite Vulkan callers who set fDeviceLostHandler now receive Skia's full device-lost payload on the driver-owned thread — description + address-info array + vendor-info array + vendor binary blob (the last three empty when VK_EXT_device_fault isn't enabled).
• Callers who leave fDeviceLostHandler null get today's behavior (no callback).
• No behavior change for existing null-slot callers.

Testing

Verified in the companion SkiaSharp PR: 5 proxy-level marshalling tests + 5 Vulkan integration tests. All pass locally on Linux/Lavapipe alongside the full Vulkan regression suite (28 pass / 2 skip / 0 fail).

Actually forcing VK_ERROR_DEVICE_LOST in CI isn't realistic (needs hardware TDR); the proxy tests cover the marshalling, the integration tests cover the wiring.

Checklist

• Targets the skiasharp branch
• Changes above lists every added/changed C API export (or "None.")
• Companion mono/SkiaSharp PR linked above (submodule bump at minimum; regenerates bindings + adds tests for C API changes)