Add Semi-SL MeanTeacher for semantic segmentation

[kprokofi]

Summary

This PR intends to add Semi-SL functionality for semantic segmentation task. This PR doesn't consider refactoring changes.

• Add MeanTeacher algorithm
• Semi-SL works for semantic segmentation
• Conduct experiments. Results are better then supervised training at least on few percent
• Add unit tests, integration and performance

In the next PRs the following changes are expected:

• Semantic Segmentation refactoring including aligning models with TrainType introduced by @harimkang. This PR doesn't include this change and simply add new segmentation model with semi_sl suffix
• Prototype based approach as in OTX1.6. Additional experiments are required

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have ran e2e tests and there is no issues.
• I have added the description of my changes into CHANGELOG in my target branch (e.g., CHANGELOG in develop).​
• I have updated the documentation in my target branch accordingly (e.g., documentation in develop).
• I have linked related issues.

License

• I submit my code changes under the same Apache License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2024 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

[kprokofi]

Pascal-VOC results:

<style> </style>

	Supervised	Semi-Supervised
all	0.6595	0.7192
dino_v2	0.8634	0.8768
litehrnet_18	0.5025	0.5978
litehrnet_s	0.4241	0.5376
litehrnet_x	0.3675	0.5185
segnext_b	0.8475	0.8520
segnext_s	0.8171	0.8409
segnext_t	0.7942	0.8109

DIS5K + Cityscapes results

<style> </style>

	Supervised	Semi-Supervised
all	0.7213	0.7327
dino_v2	0.7875	0.7979
litehrnet_18	0.6627	0.6802
litehrnet_s	0.6389	0.6605
litehrnet_x	0.6930	0.7107
segnext_b	0.7759	0.7627
segnext_s	0.7571	0.7825
segnext_t	0.7337	0.7345

The results were averaged across 5 runs with different seeds (OTX perf tests)

Unfortunately, there is no always good result. For example, when labeled images are only 6-20, the pseudo-labels are not reliable and we can observe degradation in metric. Some other approaches / additional fine-tuning is required in that case:

<style> </style>

	Supervised	Semi-Supervised
all	0.7693	0.7267
dino_v2	0.8560	0.8319
litehrnet_18	0.7548	0.7540
litehrnet_s	0.7261	0.6988
litehrnet_x	0.7378	0.7343
segnext_b	0.7805	0.6942
segnext_s	0.7718	0.7293
segnext_t	0.7584	0.6445