use nms in the model itself, simplify and make postprocess faster.

make inference in batches, fix c# handling, add overlap handling

Azaion.Inference/annotation.pyx

@@ -1,7 +1,9 @@
 import msgpack
+from pathlib import Path
 
 cdef class Detection:
     def __init__(self, double x, double y, double w, double h, int cls, double confidence):
+        self.annotation_name = None
         self.x = x
         self.y = y
         self.w = w
@@ -12,18 +14,44 @@ cdef class Detection:
     def __str__(self):
         return f'{self.cls}: {self.x:.2f} {self.y:.2f} {self.w:.2f} {self.h:.2f}, prob: {(self.confidence*100):.1f}%'
 
+    cdef overlaps(self, Detection det2):
+        cdef double overlap_x = 0.5 * (self.w + det2.w) - abs(self.x - det2.x)
+        cdef double overlap_y = 0.5 * (self.h + det2.h) - abs(self.y - det2.y)
+        cdef double overlap_area = max(0.0, overlap_x) * max(0.0, overlap_y)
+        cdef double min_area = min(self.w * self.h, det2.w * det2.h)
+
+        return overlap_area / min_area > 0.6
+
 cdef class Annotation:
-    def __init__(self, long time, list[Detection] detections):
-        self.time = time
+    def __init__(self, str name, long ms, list[Detection] detections):
+        self.original_media_name = Path(<str>name).stem.replace(" ", "")
+        self.name = f'{self.original_media_name}_{self.format_time(ms)}'
+        self.time = ms
         self.detections = detections if detections is not None else []
+        for d in self.detections:
+            d.annotation_name = self.name
         self.image = b''
 
+    cdef format_time(self, ms):
+        # Calculate hours, minutes, seconds, and hundreds of milliseconds.
+        h = ms // 3600000  # Total full hours.
+        ms_remaining = ms % 3600000
+        m = ms_remaining // 60000  # Full minutes.
+        ms_remaining %= 60000
+        s = ms_remaining // 1000  # Full seconds.
+        f = (ms_remaining % 1000) // 100  # Hundreds of milliseconds.
+        h = h % 10
+        return f"{h}{m:02}{s:02}{f}"
+
     cdef bytes serialize(self):
         return msgpack.packb({
+            "n": self.name,
+            "mn": self.original_media_name,
             "i": self.image,  # "i" = image
             "t": self.time,  # "t" = time
             "d": [  # "d" = detections
                 {
+                    "an": det.annotation_name,
                     "x": det.x,
                     "y": det.y,
                     "w": det.w,