fix: separate postprocessing from runtime inference logic and handle null masks results

[CuriousDolphin]

• Move postprocessing functions from runtime.py to utils/vision.py
• Update runtime classes to return raw model outputs
• Modify local and remote model classes to apply postprocessing separately
• Improve logging with more detailed inference time and detection count information
• Enhance error handling for cases with no detections (empty masks)

Context

New segmentation postprocess is broken when there are empty masks (non_null_pixel =0 or mask.shape !=2) in model output.

[github-actions]

Coverage Report •

File	Stmts	Miss	Cover	Missing
local_model.py	68	2	97%	143–144
remote_model.py	143	39	72%	228–233, 238, 242–243, 245–246, 250–251, 279–287, 289–292, 296–298, 302, 305–310, 312–313, 370
runtime.py	140	95	32%	30, 38–39, 60, 64, 68, 78, 80–81, 83–85, 88–90, 93, 96–98, 101, 104–105, 108–109, 112, 137–141, 143–144, 147–150, 152–153, 155, 159–162, 166–167, 169–171, 173–177, 179–180, 182–184, 188, 198–199, 209–212, 214, 216–217, 219–225, 229–232, 236–237, 239–240, 242–246, 248–249, 251–252, 254, 264–265
utils
vision.py	141	5	96%	124, 162, 192, 330, 362
TOTAL	1089	156	85%

Tests	Skipped	Failures	Errors	Time
116	1 💤	0 ❌	0 🔥	14.027s ⏱️