Fix startup crash on Vulkan devices reporting a huge maxUniformBufferRange.

[Hparty]

Problem

On some Vulkan devices (e.g. OPPO X9, Android 16), the app crashes at startup:

VulkanBuffer::Make() vmaCreateBuffer failed: VK_ERROR_OUT_OF_DEVICE_MEMORY
[GlobalCache::findOrCreateUniformBuffer] failed to create initial uniform buffer, request buffer size: 96

GlobalCache::findOrCreateUniformBuffer used std::max(device maxUBOSize, 64KB) as the pool allocation size. Many mobile GPUs report a maxUniformBufferRange of 2GB~4GB (uint32), so tgfx tried to allocate a multi-GB uniform buffer at startup and ran out of device memory. The value was also narrowed to int, which could overflow.

Close #1509

Fix

Decouple the two roles the device limit used to play, matching the behavior verified by Skia's DrawBufferManager:

• Single request limit: keeps the original std::max(device, 64KB) semantics, used only to validate one request's bind range.
• Pool allocation size: fixed at MAX_UNIFORM_BUFFER_SIZE (64KB), no longer follows the device limit. The pool grows with additional 64KB buffers on demand as before.
• Requests larger than the pool block now get a dedicated buffer instead of being rejected.
• GPULimits::maxUniformBufferBindingSize and ShaderCaps::maxUBOSize are now int64_t so Vulkan's uint32 limit is stored without overflow.

Verification

• TGFXFullTest_OpenGL: 591/591 passed on macOS.
• Vulkan configuration compiles (running Vulkan tests requires a device with a Vulkan loader).