Implemented v9-t model variant

[cansik]

I have implemented the tiny (v9-t) model variant of the v9 model, based on the original v9-s model description. I haven't trained it on COCO but on a custom dataset, which appeared to perform as expected.

Here are the model statistics (for single-class training):

┏━━━┳━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃   ┃ Name   ┃ Type                 ┃ Params ┃ Mode  ┃
┡━━━╇━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ 0 │ model  │ YOLO                 │  2.7 M │ train │
│ 1 │ metric │ MeanAveragePrecision │      0 │ train │
│ 2 │ ema    │ YOLO                 │  2.7 M │ train │
└───┴────────┴──────────────────────┴────────┴───────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Attributes                             ┃ Value      ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Trainable params                       │ 2.7 M      │
│ Non-trainable params                   │ 2.7 M      │
│ Total params                           │ 5.3 M      │
│ Total estimated model params size (MB) │ 21         │
│ Modules in train mode                  │ 2967       │
│ Modules in eval mode                   │ 0          │
└────────────────────────────────────────┴────────────┘

Related to #92

[zsp1987]

@cansik do you have a trained model?

[cansik]

@zsp1987 Not trained on COCO, I used it for a custom dataset.