refactor: decompose PatchInferenceEngine.predict into pipelined stages (#69)

Split predict() into _prepare_subject, _run_inference, and _postprocess
for better modularity. Add _PreparedSubject dataclass and _save_prediction
helper. Switch from torch.no_grad to torch.inference_mode for faster inference.

Author: skaliy

fastMONAI/_modidx.py

@@ -553,10 +553,18 @@
                                                                                          'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch.PatchInferenceEngine.__init__': ( 'vision_patch.html#patchinferenceengine.__init__',
                                                                                                   'fastMONAI/vision_patch.py'),
+                                        'fastMONAI.vision_patch.PatchInferenceEngine._postprocess': ( 'vision_patch.html#patchinferenceengine._postprocess',
+                                                                                                      'fastMONAI/vision_patch.py'),
+                                        'fastMONAI.vision_patch.PatchInferenceEngine._prepare_subject': ( 'vision_patch.html#patchinferenceengine._prepare_subject',
+                                                                                                          'fastMONAI/vision_patch.py'),
+                                        'fastMONAI.vision_patch.PatchInferenceEngine._run_inference': ( 'vision_patch.html#patchinferenceengine._run_inference',
+                                                                                                        'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch.PatchInferenceEngine.predict': ( 'vision_patch.html#patchinferenceengine.predict',
                                                                                                  'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch.PatchInferenceEngine.to': ( 'vision_patch.html#patchinferenceengine.to',
                                                                                             'fastMONAI/vision_patch.py'),
+                                        'fastMONAI.vision_patch._PreparedSubject': ( 'vision_patch.html#_preparedsubject',
+                                                                                     'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch._extract_tio_transform': ( 'vision_patch.html#_extract_tio_transform',
                                                                                            'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch._get_default_device': ( 'vision_patch.html#_get_default_device',
@@ -565,6 +573,8 @@
                                                                                              'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch._predict_patch_tta': ( 'vision_patch.html#_predict_patch_tta',
                                                                                        'fastMONAI/vision_patch.py'),
+                                        'fastMONAI.vision_patch._save_prediction': ( 'vision_patch.html#_save_prediction',
+                                                                                     'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch._warn_config_override': ( 'vision_patch.html#_warn_config_override',
                                                                                           'fastMONAI/vision_patch.py'),
                                         'fastMONAI.vision_patch.create_patch_sampler': ( 'vision_patch.html#create_patch_sampler',