fix(schema): expose gpu fp32_tflops in API responses

fp32_tflops lived on the GPU model and ingest path but was never added to the
GPURead response schema or gpu_read serializer, so the API silently dropped it.
Wire it through so the computed peak-FP32 values surface in /v1/gpus.

Author: Seungpyo1007

app/schemas/gpu.py

@@ -35,6 +35,7 @@ class GPURead(BaseModel):
     boost_clock_mhz: int
     tdp_w: int
     pcie_version: str
+    fp32_tflops: float | None = None
     blender_score: float | None = None
     verified: bool
     source_urls: list[str]

app/schemas/serializers.py

@@ -126,6 +126,7 @@ def gpu_read(gpu: DiscreteGPU, manufacturer: Brand) -> GPURead:
         boost_clock_mhz=gpu.boost_clock_mhz,
         tdp_w=gpu.tdp_w,
         pcie_version=gpu.pcie_version,
+        fp32_tflops=gpu.fp32_tflops,
         blender_score=gpu.blender_score,
         verified=gpu.verified,
         source_urls=gpu.source_urls,